研究者データベース

島崎 秀昭(シマザキ ヒデアキ)
人間知・脳・AI研究教育センター
特任准教授

基本情報

所属

  • 人間知・脳・AI研究教育センター

職名

  • 特任准教授

学位

  • Ph.D.(2007年03月 京都大学)
  • MA(2003年11月 ジョンズ・ホプキンス大学)

ORCID ID

J-Global ID

プロフィール

  • 北海道大学 人間知・脳・AI研究教育センター 特任准教授(2020- ) 京都大学大学院情報学研究科知能情報学専攻 特定准教授(2017-2020) ホンダ・リサーチ・インスティチュート・ジャパン シニアサイエンティスト(2016-2020) 理化学研究所脳科学総合研究センター研究員(2011-2016) マサチューセッツ工科大学訪問研究員(2009-2010) 理化学研究所脳科学総合研究センター訪問研究員(2007-2010) 日本学術振興会特別研究員(2006-2008, 2008-) 京都大学理学研究科物理学・宇宙物理学専攻博士(理学)(2007) ジョンズ・ホプキンス大学医学部神経科学科修士(神経科学)(2003) 慶應義塾大学理工学部物理情報工学科学士(工学)(2000)

研究キーワード

  • 情報理論   機械学習   統計力学   神経符号化   

研究分野

  • ライフサイエンス / 基盤脳科学
  • 自然科学一般 / 応用数学、統計数学
  • 情報通信 / 統計科学

職歴

  • 2020年04月 - 現在 北海道大学 人間知・脳・AI研究教育センター 特任准教授
  • 2017年04月 - 2020年03月 京都大学 大学院情報学研究科 知能情報学専攻 特定准教授
  • 2016年04月 - 2020年03月 ホンダ・リサーチ・インスティチュート・ジャパン シニアサイエンティスト

所属学協会

  • 北米神経科学会   日本神経科学学会   日本神経回路学会   

研究活動情報

論文

  • Kazuyoshi Tatsumi, Yasuhiro Inamura, Maiko Kofu, Ryoji Kiyanagi, Hideaki Shimazaki
    Journal of Applied Crystallography 55 3 533 - 543 2022年06月01日 
    A data-driven bin-width optimization for the histograms of measured data sets based on inhomogeneous Poisson processes was developed in a neurophysiology study [Shimazaki & Shinomoto (2007). Neural Comput.19, 1503–1527], and a subsequent study [Muto, Sakamoto, Matsuura, Arima & Okada (2019). J. Phys. Soc. Jpn, 88, 044002] proposed its application to inelastic neutron scattering (INS) data. In the present study, the results of the method on experimental INS time-of-flight data collected under different measurement conditions from a copper single crystal are validated. The extrapolation of the statistics on a given data set to other data sets with different total counts precisely infers the optimal bin widths on the latter. The histograms with the optimized bin widths statistically verify two fine-spectral-feature examples in the energy and momentum transfer cross sections: (i) the existence of phonon band gaps; and (ii) the number of plural phonon branches located close to each other. This indicates that the applied method helps in the efficient and rigorous observation of spectral structures important in physics and materials science like novel forms of magnetic excitation and phonon states correlated to thermal conductivities.
  • Takuya Isomura, Hideaki Shimazaki, Karl Friston
    Communications Biology 5 55  2022年01月14日 [査読有り][通常論文]
     
    This work considers a class of canonical neural networks comprising rate coding models, wherein neural activity and plasticity minimise a common cost function—and plasticity is modulated with a certain delay. We show that such neural networks implicitly perform active inference and learning to minimise the risk associated with future outcomes. Mathematical analyses demonstrate that this biological optimisation can be cast as maximisation of model evidence, or equivalently minimisation of variational free energy, under the well-known form of a partially observed Markov decision process model. This equivalence indicates that the delayed modulation of Hebbian plasticity—accompanied with adaptation of firing thresholds—is a sufficient neuronal substrate to attain Bayes optimal control. We corroborated this proposition using numerical analyses of maze tasks. This theory offers a universal characterisation of canonical neural networks in terms of Bayesian belief updating and provides insight into the neuronal mechanisms underlying planning and adaptive behavioural control.
  • Makio Torigoe, Tanvir Islam, Hisaya Kakinuma, Chi Chung Alan Fung, Takuya Isomura, Hideaki Shimazaki, Tazu Aoki, Tomoki Fukai, Hitoshi Okamoto
    Nature Communications 10.1101/546440  Springer Nature 2021年09月29日 [査読有り][通常論文]
  • Miguel Aguilera, S. Amin Moosavi, Hideaki Shimazaki
    Nature Communications 12 1197  2021年01月19日 [査読有り][通常論文]
     
    Kinetic Ising models are powerful tools for studying the non-equilibrium dynamics of complex systems. As their behavior is not tractable for large networks, many mean-field methods have been proposed for their analysis, each based on unique assumptions about the system’s temporal evolution. This disparity of approaches makes it challenging to systematically advance mean-field methods beyond previous contributions. Here, we propose a unifying framework for mean-field theories of asymmetric kinetic Ising systems from an information geometry perspective. The framework is built on Plefka expansions of a system around a simplified model obtained by an orthogonal projection to a sub-manifold of tractable probability distributions. This view not only unifies previous methods but also allows us to develop novel methods that, in contrast with traditional approaches, preserve the system’s correlations. We show that these new methods can outperform previous ones in predicting and assessing network properties near maximally fluctuating regimes.
  • Haruna Nakajo, Ming-Yi Chou, Masae Kinoshita, Lior Appelbaum, Hideaki Shimazaki, Takashi Tsuboi, Hitoshi Okamoto
    Cell reports 31 12 107790 - 107790 2020年06月23日 [査読有り][通常論文]
     
    Many animals fight for dominance between conspecifics. Because winners could obtain more resources than losers, fighting outcomes are important for the animal's survival, especially in a situation with insufficient resources, such as hunger. However, it remains unclear whether and how hunger affects fighting outcomes. Herein, we investigate the effects of food deprivation on brain activity and fighting behaviors in zebrafish. We report that starvation induces winning in social conflicts. Before the fights, starved fish show potentiation of the lateral subregion of the dorsal habenula (dHbL)-dorsal/intermediate interpeduncular nucleus (d/iIPN) pathway, which is known to be essential for and potentiated after winning fights. Circuit potentiation is mediated by hypothalamic orexin/hypocretin neuropeptides, which prolong AMPA-type glutamate receptor (AMPAR) activity by increasing the expression of a flip type of alternative splicing variant of the AMPAR subunit. This mechanism may underlie how hungry vertebrates win fights and may be commonly shared across animal phylogeny.
  • ベイズ統計と熱力学から見る 生物の学習と認識のダイナミクス
    島崎 秀昭
    日本神経回路学会誌 26 3 72 - 98 2019年09月 [査読無し][通常論文]
     
    本稿は生物の学習と認識をベイズ統計学と熱力学の立場から解説する.脳は学習により獲得 した外界のモデルをもとに推論を行う器官であると考えるのが脳のベイズ的な見方である.本 稿の前半はこの見方が自発活動と刺激誘起活動をめぐる実験事実からいかにして形成されてき たかを解説し,ベイズ推論に基づく外界の認識が実現されていることを示唆する神経活動とし て図地分離・注意等の実験例を紹介する.後者の実験では順方向結合による初期刺激応答に対 してリカレント回路による入力が時間遅れで統合されていることが示唆され,初期刺激応答後 の神経活動に気づき・注意・報酬価値が表されると報告している.後半では神経細胞集団の刺 激応答のダイナミクスを生成モデルを用いて再現した上で,学習と認識を神経活動のエントロ ピーに関する法則を用いて記述する熱力学的手法について解説する.特に時間遅れを伴う動的 な情報統合によってベイズ推論を実現する神経ダイナミクスが情報論的なエンジンを構成する ことを紹介する(ニューラルエンジン).これにより内発的な刺激変調を定量化し,その効率 を計算できることを解説する.
  • Jimmy Gaudreault, Arunabh Saxena, Hideaki Shimazaki
    2019 International Joint Conference on Neural Networks (IJCNN) 1 - 8 2019年01月 [査読有り][通常論文]
     
    Many time-series data including text, movies, and biological signals can be represented as sequences of correlated binary patterns. These patterns may be described by weighted combinations of a few dominant structures that underpin specific interactions among the binary elements. To extract the dominant correlation structures and their contributions to generating data in a time-dependent manner, we model the dynamics of binary patterns using the state-space model of an Ising-type network that is composed of multiple undirected graphs. We provide a sequential Bayes algorithm to estimate the dynamics of weights on the graphs while gaining the graph structures online. This model can uncover overlapping graphs underlying the data better than a traditional orthogonal decomposition method, and outperforms an original time-dependent full Ising model. We assess the performance of the method by simulated data, and demonstrate that spontaneous activity of cultured hippocampal neurons is represented by dynamics of multiple graphs.
  • Jimmy Gaudreault, Hideaki Shimazaki
    Lecture Notes in Computer Science 11141 641 - 651 2018年10月 [査読有り][通常論文]
     
    In this study, we analyzed the activity of monkey V1 neu- rons responding to grating stimuli of different orientations using inference methods for a time-dependent Ising model. The method provides optimal estimation of time-dependent neural interactions with credible intervals according to the sequential Bayes estimation algorithm. Furthermore, it allows us to trace dynamics of macroscopic network properties such as entropy, sparseness, and fluctuation. Here we report that, in all exam- ined stimulus conditions, pairwise interactions contribute to increasing sparseness and fluctuation. We then demonstrate that the orientation of the grating stimulus is in part encoded in the pairwise interactions of the neural populations. These results demonstrate the utility of the state- space Ising model in assessing contributions of neural interactions during stimulus processing.
  • 島崎 秀昭
    日本神経回路学会誌 25 3 86 - 103 2018年09月 [査読無し][通常論文]
     
    本稿は生物の認識と行動の環境への適応に関する理論の解説を行う.始めに環境の階層モデルに対する近似推論法を紹介し,サプライズ最小化の原理から生物の認識及び行動の説明を試みるFristonらの自由エネルギー原理を機械学習の立場から概観する.次に階層モデルによるアプローチを情報理論の観点から考察し,情報量最大化原理に基づく古典的な知覚の理論や非線形信号処理との関係を解説する.最後に認識のモデルに対する熱力学的な取り扱いを紹介し,学習過程のエントロピー・自由エネルギーのダイナミクスを考察して熱力学法則との形式的関係を明らかにする.このようにして生物がその非線形な演算装置を適応させて外界の表現を獲得し,身体を使い認識に沿うように外界を改変していく過程が複数の視点から明らかになる.
  • Safura Rashid Shomali, Majid Nili Ahmadabadi, Hideaki Shimazaki, Seyyed Nader Rasuli
    Journal of Computational Neuroscience 44 2 147 - 171 2018年04月01日 [査読有り][通常論文]
     
    The noisy threshold regime, where even a small set of presynaptic neurons can significantly affect postsynaptic spike-timing, is suggested as a key requisite for computation in neurons with high variability. It also has been proposed that signals under the noisy conditions are successfully transferred by a few strong synapses and/or by an assembly of nearly synchronous synaptic activities. We analytically investigate the impact of a transient signaling input on a leaky integrate-and-fire postsynaptic neuron that receives background noise near the threshold regime. The signaling input models a single strong synapse or a set of synchronous synapses, while the background noise represents a lot of weak synapses. We find an analytic solution that explains how the first-passage time (ISI) density is changed by transient signaling input. The analysis allows us to connect properties of the signaling input like spike timing and amplitude with postsynaptic first-passage time density in a noisy environment. Based on the analytic solution, we calculate the Fisher information with respect to the signaling input’s amplitude. For a wide range of amplitudes, we observe a non-monotonic behavior for the Fisher information as a function of background noise. Moreover, Fisher information non-trivially depends on the signaling input’s amplitude changing the amplitude, we observe one maximum in the high level of the background noise. The single maximum splits into two maximums in the low noise regime. This finding demonstrates the benefit of the analytic solution in investigating signal transfer by neurons.
  • Robert E. Kass, Shun-Ichi Amari, Kensuke Arai, Emery N. Brown, Casey O. Diekman, Markus Diesmann, Brent Doiron, Uri T. Eden, Adrienne L. Fairhall, Grant M. Fiddyment, Tomoki Fukai, Sonja Grün, Matthew T. Harrison, Moritz Helias, Hiroyuki Nakahara, Jun-Nosuke Teramae, Peter J. Thomas, Mark Reimers, Jordan Rodu, Horacio G. Rotstein, Eric Shea-Brown, Hideaki Shimazaki, Shigeru Shinomoto, Byron M. Yu, Mark A. Kramer
    Annual Review of Statistics and Its Application 5 183 - 214 2018年03月07日 [査読有り][通常論文]
     
    Mathematical and statistical models have played important roles in neuroscience, especially by describing the electrical activity of neurons recorded individually, or collectively across large networks. As the field moves forward rapidly, new challenges are emerging. For maximal effectiveness, those working to advance computational neuroscience will need to appreciate and exploit the complementary strengths of mechanistic theory and the statistical paradigm.
  • HaDi MaBouDi, Hideaki Shimazaki, Martin Giurfa, Lars Chittka
    PLOS COMPUTATIONAL BIOLOGY 13 6 e1005551  2017年06月 [査読有り][通常論文]
     
    The honeybee olfactory system is a well-established model for understanding functional mechanisms of learning and memory. Olfactory stimuli are first processed in the antennal lobe, and then transferred to the mushroom body and lateral horn through dual pathways termed medial and lateral antennal lobe tracts (m-ALT and l-ALT). Recent studies reported that honeybees can perform elemental learning by associating an odour with a reward signal even after lesions in m-ALT or blocking the mushroom bodies. To test the hypothesis that the lateral pathway (l-ALT) is sufficient for elemental learning, we modelled local computation within glomeruli in antennal lobes with axons of projection neurons connecting to a decision neuron (LHN) in the lateral horn. We show that inhibitory spike-timing dependent plasticity (modelling non-associative plasticity by exposure to different stimuli) in the synapses from local neurons to projection neurons decorrelates the projection neurons' outputs. The strength of the decorrelations is regulated by global inhibitory feedback within antennal lobes to the projection neurons. By additionally modelling octopaminergic modification of synaptic plasticity among local neurons in the antennal lobes and projection neurons to LHN connections, the model can discriminate and generalize olfactory stimuli. Although positive patterning can be accounted for by the l-ALT model, negative patterning requires further processing and mushroom body circuits. Thus, our model explains several-but not all-types of associative olfactory learning and generalization by a few neural layers of odour processing in the l-ALT. As an outcome of the combination between non-associative and associative learning, the modelling approach allows us to link changes in structural organization of honeybees' antennal lobes with their behavioural performances over the course of their life.
  • Christian Donner, Klaus Obermayer, Hideaki Shimazaki
    PLOS COMPUTATIONAL BIOLOGY 13 1 e1005309  2017年01月 [査読有り][通常論文]
     
    The models in statistical physics such as an Ising model offer a convenient way to characterize stationary activity of neural populations. Such stationary activity of neurons may be expected for recordings from in vitro slices or anesthetized animals. However, modeling activity of cortical circuitries of awake animals has been more challenging because both spike-rates and interactions can change according to sensory stimulation, behavior, or an internal state of the brain. Previous approaches modeling the dynamics of neural interactions suffer from computational cost; therefore, its application was limited to only a dozen neurons. Here by introducing multiple analytic approximation methods to a state-space model of neural population activity, we make it possible to estimate dynamic pairwise interactions of up to 60 neurons. More specifically, we applied the pseudolikelihood approximation to the state-space model, and combined it with the Bethe or TAP mean-field approximation to make the sequential Bayesian estimation of the model parameters possible. The large-scale analysis allows us to investigate dynamics of macroscopic properties of neural circuitries underlying stimulus processing and behavior. We show that the model accurately estimates dynamics of network properties such as sparseness, entropy, and heat capacity by simulated data, and demonstrate utilities of these measures by analyzing activity of monkey V4 neurons as well as a simulated balanced network of spiking neurons.
  • Yasuhiro Mochizuki, Tomokatsu Onaga, Hideaki Shimazaki, Takeaki Shimokawa, Yasuhiro Tsubo, Rie Kimura, Akiko Saiki, Yutaka Sakai, Yoshikazu Isomura, Shigeyoshi Fujisawa, Ken-ichi Shibata, Daichi Hirai, Takahiro Furuta, Takeshi Kaneko, Susumu Takahashi, Tomoaki Nakazono, Seiya Ishino, Yoshio Sakurai, Takashi Kitsukawa, Jong Won Lee, Hyunjung Lee, Min Whan Jung, Cecilia Babul, Pedro E. Maldonado, Kazutaka Takahashi, Fritzie I. Arce-McShane, Callum F. Ross, Barry J. Sessle, Nicholas G. Hatsopoulos, Thomas Brochier, Alexa Riehle, Paul Chorley, Sonja Gruen, Hisao Nishijo, Satoe Ichihara-Takeda, Shintaro Funahashi, Keisetsu Shima, Hajime Mushiake, Yukako Yamane, Hiroshi Tamura, Ichiro Fujita, Naoko Inaba, Kenji Kawano, Sergei Kurkin, Kikuro Fukushima, Kiyoshi Kurata, Masato Taira, Ken-Ichiro Tsutsui, Tadashi Ogawa, Hidehiko Komatsu, Kowa Koida, Keisuke Toyama, Barry J. Richmond, Shigeru Shinomoto
    JOURNAL OF NEUROSCIENCE 36 21 5736 - 5747 2016年05月 [査読有り][通常論文]
     
    The architectonic subdivisions of the brain are believed to be functional modules, each processing parts of global functions. Previously, we showed that neurons in different regions operate in different firing regimes in monkeys. It is possible that firing regimes reflect differences in underlying information processing, and consequently the firing regimes in homologous regions across animal species might be similar. We analyzed neuronal spike trains recorded from behaving mice, rats, cats, and monkeys. The firing regularity differed systematically, with differences across regions in one species being greater than the differences in similar areas across species. Neuronal firing was consistently most regular in motor areas, nearly random in visual and prefrontal/medial prefrontal cortical areas, and bursting in the hippocampus in all animals examined. This suggests that firing regularity (or irregularity) plays a key role in neural computation in each functional subdivision, depending on the types of information being carried.
  • Ming-Yi Chou, Ryunosuke Amo, Masae Kinoshita, Bor-Wei Cherng, Hideaki Shimazaki, Masakazu Agetsuma, Toshiyuki Shiraki, Tazu Aoki, Mikako Takahoko, Masako Yamazaki, Shin-ichi Higashijima, Hitoshi Okamoto
    SCIENCE 352 6281 87 - 90 2016年04月 [査読有り][通常論文]
     
    When animals encounter conflict they initiate and escalate aggression to establish and maintain a social hierarchy. The neural mechanisms by which animals resolve fighting behaviors to determine such social hierarchies remain unknown. We identified two subregions of the dorsal habenula (dHb) in zebrafish that antagonistically regulate the outcome of conflict. The losing experience reduced neural transmission in the lateral subregion of dHb (dHbL)-dorsal/intermediate interpeduncular nucleus (d/iIPN) circuit. Silencing of the dHbL or medial subregion of dHb (dHbM) caused a stronger predisposition to lose or win a fight, respectively. These results demonstrate that the dHbL and dHbM comprise a dual control system for conflict resolution of social aggression.
  • HaDi MaBouDi, Hideaki Shimazaki, Shun-ichi Amari, Hamid Soltanian-Zadeh
    VISION RESEARCH 120 61 - 73 2016年03月 [査読有り][通常論文]
     
    Natural scenes contain richer perceptual information in their spatial phase structure than their amplitudes. Modeling phase structure of natural scenes may explain higher-order structure inherent to the natural scenes, which is neglected in most classical models of redundancy reduction. Only recently, a few models have represented images using a complex form of receptive fields (RFs) and analyze their complex responses in terms of amplitude and phase. However, these complex representation models often tacitly assume a uniform phase distribution without empirical support. The structure of spatial phase distributions of natural scenes in the form of relative contributions of paired responses of RFs in quadrature has not been explored statistically until now. Here, we investigate the spatial phase structure of natural scenes using complex forms of various Gabor-like RFs. To analyze distributions of the spatial phase responses, we constructed a mixture model that accounts for multi-modal circular distributions, and the EM algorithm for estimation of the model parameters. Based on the likelihood, we report presence of both uniform and structured bimodal phase distributions in natural scenes. The latter bimodal distributions were symmetric with two peaks separated by about 180 degrees. Thus, the redundancy in the natural scenes can be further removed by using the bimodal phase distributions obtained from these RFs in the complex representation models. These results predict that both phase invariant and phase sensitive complex cells are required to represent the regularities of natural scenes in visual systems. (C) 2015 Elsevier Ltd. All rights reserved.
  • Christian Donner, Hideaki Shimazaki
    NEURAL INFORMATION PROCESSING, ICONIP 2016, PT III 9949 104 - 110 2016年 [査読有り][通常論文]
     
    The maximum entropy method has been successfully employed to explain stationary spiking activity of a neural population by using fewer features than the number of possible activity patterns. Modeling network activity in vivo, however, has been challenging because features such as spike-rates and interactions can change according to sensory stimulation, behavior, or brain state. To capture the time-dependent activity, Shimazaki et al. (PLOS Comp Biol, 2012) previously introduced a state-space framework for the latent dynamics of neural interactions. However, the exact method suffers from computational cost; therefore its application was limited to only similar to 15 neurons. Here we introduce the pseudolikelihood method combined with the TAP or Bethe approximation to the state-space model, and make it possible to estimate dynamic pairwise interactions of up to 30 neurons. These analytic approximations allow analyses of time-varying activity of larger networks in relation to stimuli or behavior.
  • Hideaki Shimazaki
    arXiv 1512.07855  2015年12月 [査読無し][通常論文]
     
    We show that dynamical gain modulation of neurons' stimulus response is described as an information-theoretic cycle that generates entropy associated with the stimulus-related activity from entropy produced by the modulation. To articulate this theory, we describe stimulus-evoked activity of a neural population based on the maximum entropy principle with constraints on two types of overlapping activities, one that is controlled by stimulus conditions and the other, termed internal activity, that is regulated internally in an organism. We demonstrate that modulation of the internal activity realises gain control of stimulus response, and controls stimulus information. A cycle of neural dynamics is then introduced to model information processing by the neurons during which the stimulus information is dynamically enhanced by the internal gain-modulation mechanism. Based on the conservation law for entropy production, we demonstrate that the cycle generates entropy ascribed to the stimulus-related activity using entropy supplied by the internal mechanism, analogously to a heat engine that produces work from heat. We provide an efficient cycle that achieves the highest entropic efficiency to retain the stimulus information. The theory allows us to quantify efficiency of the internal computation and its theoretical limit.
  • Hideaki Shimazaki, Kolia Sadeghi, Tomoe Ishikawa, Yuji Ikegaya, Taro Toyoizumi
    Scientific Reports 5 9821  2015年04月 [査読有り][通常論文]
     
    Activity patterns of neural population are constrained by underlying biological mechanisms. These patterns are characterized not only by individual activity rates and pairwise correlations but also by statistical dependencies among groups of neurons larger than two, known as higher-order interactions (HOIs). While HOIs are ubiquitous in neural activity, primary characteristics of HOIs remain unknown. Here, we report that simultaneous silence (SS) of neurons concisely summarizes neural HOIs. Spontaneously active neurons in cultured hippocampal slices express SS that is more frequent than predicted by their individual activity rates and pairwise correlations. The SS explains structured HOIs seen in the data, namely, alternating signs at successive interaction orders. Inhibitory neurons are necessary to maintain significant SS. The structured HOIs predicted by SS were observed in a simple neural population model characterized by spiking nonlinearity and correlated input. These results suggest that SS is a ubiquitous feature of HOIs that constrain neural activity patterns and can influence information processing.
  • Hideaki Shimazaki
    Journal of Physics: Conference Series 473 1 012009  2013年 [査読有り][招待有り]
     
    Neurons in cortical circuits exhibit coordinated spiking activity, and can produce correlated synchronous spikes during behavior and cognition. We recently developed a method for estimating the dynamics of correlated ensemble activity by combining a model of simultaneous neuronal interactions (e.g., a spin-glass model) with a state-space method (Shimazaki et al. 2012 PLoS Comput Biol 8 e1002385). This method allows us to estimate stimulus-evoked dynamics of neuronal interactions which is reproducible in repeated trials under identical experimental conditions. However, the method may not be suitable for detecting stimulus responses if the neuronal dynamics exhibits significant variability across trials. In addition, the previous model does not include effects of past spiking activity of the neurons on the current state of ensemble activity. In this study, we develop a parametric method for simultaneously estimating the stimulus and spike-history effects on the ensemble activity from single-trial data even if the neurons exhibit dynamics that is largely unrelated to these effects. For this goal, we model ensemble neuronal activity as a latent process and include the stimulus and spike-history effects as exogenous inputs to the latent process. We develop an expectation-maximization algorithm that simultaneously achieves estimation of the latent process, stimulus responses, and spike-history effects. The proposed method is useful to analyze an interaction of internal cortical states and sensory evoked activity. © Published under licence by IOP Publishing Ltd.
  • Hideaki Shimazaki, Shun-ichi Amari, Emery N. Brown, Sonja Gruen
    PLOS COMPUTATIONAL BIOLOGY 8 3 e1002385  2012年03月 [査読有り][通常論文]
     
    Precise spike coordination between the spiking activities of multiple neurons is suggested as an indication of coordinated network activity in active cell assemblies. Spike correlation analysis aims to identify such cooperative network activity by detecting excess spike synchrony in simultaneously recorded multiple neural spike sequences. Cooperative activity is expected to organize dynamically during behavior and cognition; therefore currently available analysis techniques must be extended to enable the estimation of multiple time-varying spike interactions between neurons simultaneously. In particular, new methods must take advantage of the simultaneous observations of multiple neurons by addressing their higher-order dependencies, which cannot be revealed by pairwise analyses alone. In this paper, we develop a method for estimating time-varying spike interactions by means of a state-space analysis. Discretized parallel spike sequences are modeled as multi-variate binary processes using a log-linear model that provides a well-defined measure of higher-order spike correlation in an information geometry framework. We construct a recursive Bayesian filter/smoother for the extraction of spike interaction parameters. This method can simultaneously estimate the dynamic pairwise spike interactions of multiple single neurons, thereby extending the Ising/spin-glass model analysis of multiple neural spike train data to a nonstationary analysis. Furthermore, the method can estimate dynamic higher-order spike interactions. To validate the inclusion of the higher-order terms in the model, we construct an approximation method to assess the goodness-of-fit to spike data. In addition, we formulate a test method for the presence of higher-order spike correlation even in nonstationary spike data, e. g., data from awake behaving animals. The utility of the proposed methods is tested using simulated spike data with known underlying correlation dynamics. Finally, we apply the methods to neural spike data simultaneously recorded from the motor cortex of an awake monkey and demonstrate that the higher-order spike correlation organizes dynamically in relation to a behavioral demand.
  • Hideaki Shimazaki, Shigeru Shinomoto
    JOURNAL OF COMPUTATIONAL NEUROSCIENCE 29 1-2 171 - 182 2010年08月 [査読有り][通常論文]
     
    Kernel smoother and a time-histogram are classical tools for estimating an instantaneous rate of spike occurrences. We recently established a method for selecting the bin width of the time-histogram, based on the principle of minimizing the mean integrated square error (MISE) between the estimated rate and unknown underlying rate. Here we apply the same optimization principle to the kernel density estimation in selecting the width or "bandwidth" of the kernel, and further extend the algorithm to allow a variable bandwidth, in conformity with data. The variable kernel has the potential to accurately grasp non-stationary phenomena, such as abrupt changes in the firing rate, which we often encounter in neuroscience. In order to avoid possible overfitting that may take place due to excessive freedom, we introduced a stiffness constant for bandwidth variability. Our method automatically adjusts the stiffness constant, thereby adapting to the entire set of spike data. It is revealed that the classical kernel smoother may exhibit goodness-of-fit comparable to, or even better than, that of modern sophisticated rate estimation methods, provided that the bandwidth is selected properly for a given set of spike data, according to the optimization methods presented here.
  • Hideaki Shimazaki, Shun-ichi Amari, Emery N. Brown, Sonja Gruen
    2009 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING, VOLS 1- 8, PROCEEDINGS 3501 - + 2009年 [査読有り][招待有り]
     
    A state-space method for simultaneously estimating time-dependent rate and higher-order correlation underlying parallel spike sequences is proposed. Discretized parallel spike sequences are modeled by a conditionally independent multivariate Bernoulli process using a log-linear link function, which contains a state of higher-order interaction factors. A nonlinear recursive filtering formula is derived from a log-quadratic approximation to the posterior distribution of the state. Together with a fixed-interval smoothing algorithm, time-dependent log-linear parameters are estimated. The smoothed estimates are optimized via EM-algorithm such that their prior covariance matrix maximizes the expected complete data log-likelihood. In addition, we perform model selection on the hierarchical log-linear state-space models to avoid over-fitting. Application of the method to simultaneously recorded neuronal spike sequences is expected to contribute to uncover dynamic cooperative activities of neurons in relation to behavior.
  • Hideaki Shimazaki, Shigeru Shinomoto
    NEURAL COMPUTATION 19 6 1503 - 1527 2007年06月 [査読有り][通常論文]
     
    The time histogram method is the most basic tool for capturing a time-dependent rate of neuronal spikes. Generally in the neurophysiological literature, the bin size that critically determines the goodness of the fit of the time histogram to the underlying spike rate has been subjectively selected by individual researchers. Here, we propose a method for objectively selecting the bin size from the spike count statistics alone, so that the resulting bar or line graph time histogram best represents the unknown underlying spike rate. For a small number of spike sequences generated from a modestly fluctuating rate, the optimal bin size may diverge, indicating that any time histogram is likely to capture a spurious rate. Given a paucity of data, the method presented here can nevertheless suggest how many experimental trials should be added in order to obtain a meaningful time-dependent histogram with the required accuracy.
  • A recipe for optimizing a time-histogram
    Shimazaki H, Shinomoto S
    Advances in Neural Information Processing Systems 19 1289 - 1296 2007年 [査読有り][通常論文]
  • Shimazaki Hideaki, Niebur Ernst
    Progress of theoretical physics. Supplement 161 336 - 339 Published for the Yukawa Institute for Theoretical Physics and the Physical Society of Japan 2006年 
    We develop an oscillator model of selective attention based on spread spectrum communication. Stimulus intensity and attention are differentially encoded in the time average of phase velocities of sender oscillators and their temporal structure respectively. Receiver oscillators are driven by a mixture of phase velocities of the sender oscillators. With the aid of top-down modulatory signals that do not change the mean rotational velocity of sender oscillators, our proposed communication scheme allows the receiver oscillators to selectively correlate their velocity with the attended stimulus.
  • Hideaki Shimazaki, Ernst Niebur
    Physical Review E - Statistical, Nonlinear, and Soft Matter Physics 72 1 011912  2005年07月 [査読有り][通常論文]
     
    We introduce a discrete multiplicative process as a generic model of competition. Players with different abilities successively join the game and compete for finite resources. Emergence of dominant players and evolutionary development occur as a phase transition. The competitive dynamics underlying this transition is understood from a formal analogy to statistical mechanics. The theory is applicable to bacterial competition, predicting novel population dynamics near criticality. © 2005 The American Physical Society.
  • Minoru Tsukada, Takeshi Aihara, Yuki Kobayashi, Hideaki Shimazaki
    HIPPOCAMPUS 15 1 104 - 109 2005年 [査読有り][通常論文]
     
    Spike-timing-dependent long-term potentiation (LTP) and long-term depression (LTD) were investigated in the CA1 area of hippocampal slices using optical imaging. A pair of electrical pulses were used to stimulate the Schaffer-commissural collateral and the stratum oriens with various sets of relative timing (tau) between the two stimuli. These sets of paired pulses gave rise to LTP/LTD, whose induction yeas closely related to tau, and their profiles were classified into two types depending on their layer specific location along the dendrite. One was characterized by a symmetric time window observed in the proximal region of the stratum radiatum (SR) and the other by an asymmetric time window in the distal region of the SR. Bath application of bicuculline (gamma-aminobutyric acid [GABA] receptor antagonist) to hippocampal slices revealed that GABAergic interneuron projections were responsible for the symmetry of a time window. (c) 2004 Wiley-Liss, Inc.
  • 小林祐喜, 島崎秀昭, 相良威, 塚田稔
    日本神経回路学会誌 = The Brain & neural networks 8 2 57 - 64 Japanese Neural Network Society 2001年06月05日 [査読無し][通常論文]
     
    Spatial distributions of LTP/LTD were investigated in the CA1 area of hippocampal slices by optical imaging method. A pair of electrical pulse stimuli was used to stimulate the schaffer collateral (Stim. A) and the stratum oriens (Stim. B) with various sets of the relative timing (τ). LTP was induced with -10ms<τ<10ms, τ=0ms led to the widest and strongest LTP. On the other hand, τ=±20ms induced LTD. In particular, this LTD at τ=20ms was something that had not been observed by others using cultured hippocampal neurons. The LTD at τ=20ms was blocked by bicuculline (GABA (gamma-aminobutyric acid) receptor antagonist). The importance of precise spike timing in hippocampal CA1 network is discussed in the light of computational neuroscience involving the covariance learning rule.

書籍

  • 島崎秀昭 (担当:分担執筆)
    脳科学辞典編集委員会 2021年08月 
    神経符号化とは、外界の刺激が神経活動に変換・表現され、行動を担う神経活動が生成される過程を指す。刺激・行動と神経活動の関係を記述し、刺激の認識や行動生成を担う神経活動とその機構を同定することで、この過程を明らかにする研究を神経符号化研究(neural coding studies)という。同定された神経活動・機構を神経符号(neural code)と呼ぶ。
  • Dynamic Neuroscience: Statistics, Modeling, and Control
    Ed. Zhe Chen, Sridevi Sarma (担当:分担執筆範囲:Chapter 11: Neural Engine Hypothesis)
    Springer, Cham 2018年01月 (ISBN: 9783319719757) 328 267-291
  • 金澤一郎, 宮下保司, Eric R. Kandel, James H. Schwartz, Steven A. Siegelbaum, Thomas M.Jessell, A. J. Hudspeth (担当:単訳範囲:付録F:神経科学への理論的アプローチ:単一ニューロンからネットワークまで)
    金澤一郎, 宮下保司 メディカルサイエンスインターナショナル 2014年04月 (ISBN: 4895927717) 1696

講演・口頭発表等

  • 神経科学研究の理論的アプローチ  [招待講演]
    島崎秀昭
    京都大学情報学研究科システム科学専攻 集中講義 2022年09月 公開講演,セミナー,チュートリアル,講習,講義等
  • 島崎秀昭
    応用脳科学コンソーシアム アドバンスコース「脳に学ぶAI」 2022年09月 公開講演,セミナー,チュートリアル,講習,講義等 
    本講演では脳の自由エネルギー原理が形成されてきた歴史的背景を,それを支える主要な実験結果や他分野との関わりとともに紹介する.自然刺激への適応に基づく古典的な認識の理論から,外界のモデルを脳の中に持つとするベイズ脳仮説,そして環境に対する働きかけを推論の枠組みに取り入れ認識と行動の統一的理解を目指す自由エネルギー原理.これらがどのような実験事実に基づき(あるいは基づかず)構築され,情報理論・機械学習・統計物理といった他分野とどのような関わりのなかで発展してきたかを解説し,ロボティクス・AIの応用研究に今後どのように繋がっていくのか展望を述べる.
  • 神経細胞集団活動の数理とデータ解析
    島崎秀昭
    日本応用数理学会 2022年度年会 2022年09月 口頭発表(招待・特別) 
    脳の神経細胞はネットワークを形成して相互に依存しあいながら活動し,多様なダイナミクスによって外界の情報や動物の行動を符号化している.本講演では,非線形性・非定常性・非平衡性によって特徴づけられる神経細胞集団活動を明らかにする指数分布族・状態空間モデルを用いたイベント時系列データ解析方法を紹介し,活動データから背後の神経回路網の構造や神経回路網による情報表現を明らかにする試みを紹介する.
  • Hideaki Shimazaki
    International Symposium on Artificial Intelligence and Brain Science 2022 2022年07月 
    What aspects of neural activity underpin our conscious experiences? While we may need various measures to characterize them, here we propose that seeing the brain as 'an information-theoretic engine' might carve out an essential aspect of the consciousness experiences. Neurophysiological studies on early visual cortices revealed that an initial feedforward-sweep of neural response conveys stimulus features only, whereas modulation of the late component (e.g., ~100 ms after the stimulus onset) presumably mediated by feedback connections from higher brain regions carries perceptual effects such as awareness and attention. Psychophysical experiments on humans using visual masking or transcranial magnetic stimulation showed that selective disruption of the late component vanishes conscious experiences of the stimulus. Here we provide a unified computational and statistical view of the modulation of sensory representation by internal dynamics in the brain. A key computation is the nonlinear integration of multiple signals known as the gain modulation ubiquitously found in nervous systems as a mechanism to adapt neurons' nonlinear response functions to stimulus distributions. We show that the Bayesian view of the brain provides a statistical paradigm for gain modulation. More precisely, we can model the delayed gain-modulation of the stimulus-response via recurrent feedback connections as dynamics of the Bayesian inference that combines the observation and top-down prior with time delay. Interestingly, this process becomes a mathematical analog of a heat engine in thermodynamics [1,2]. This view, 'the brain as an information-theoretic engine,' allows us to quantify the amount of delayed gain modulation in terms of entropy changes in neural activity, which quantifies the perceptual capacity of neural dynamics. We present a data-driven approach to quantify the delayed gain modulation from spike data using the state-space model of neural populations [3,4]. With the theoretical and methodological advances, we share the view that the thermodynamic approaches to neural systems [4-6] will be critical to uncovering neural activity underlying conscious experiences. [1] Shimazaki (2015) Neurons as an Information-theoretic Engine. arXiv:1512.07855 (published as a book chapter) [2] Shimazaki (2020) arXiv:2006.13158 [3] Shimazaki, Amari, Brown, Gruen (2012) PLoS Comp Biol 8(3): e1002385 [4] Donner, Obermeyer, Shimazaki (2017) PLoS Comp Biol 13(1): e1005309 [5] Aguilera, Moosavi, Shimazaki (2021) Nat Commun 12, 1197 [6] Aguilera, Igarashi, Shimazaki (2022) arXiv:2205.09886
  • 状態空間キネティックイジングモデルを用いた非平衡神経ダイナミクスの解析  [通常講演]
    石原 憲, 島崎 秀昭
    NEURO2022 2022年07月 ポスター発表 沖縄 
    In neural systems, populations of neurons convey information about stimuli, decisions, and motor actions using their concerted activities. The nonequilibrium and nonstationary dynamics of the neurons hallmark the recognition and learning dynamics observed in vivo. The activity flow over asymmetric neuronal networks makes the neuronal dynamics nonequilibrium. Moreover, autonomous or external drives to the observed neurons can make their activity nonstationary, i.e., the firing rates and their interactions evolve in time. The kinetic Ising model is a powerful tool for studying the nonequilibrium neuronal dynamics, offering to model history dependency of the spiking activities, in contrast to the classical equilibrium Ising model that dictates neurons' simultaneous activities without the history effect. However, the typical kinetic Ising model assumes stationarity, limiting the models' applicability to analyze in vivo data. Here, we propose a state-space framework for the kinetic Ising model to analyze nonstationary, nonequilibrium neuronal dynamics. In this approach, the parameters of the kinetic Ising model, dictating the firing rates and their interactions, evolve in time. We developed the Bayesian filtering and smoothing algorithms to estimate the nonequilibrium neuronal dynamics and the EM algorithm to optimize the smoothness parameters of the dynamics. This framework extends the previous state-space modeling of the equilibrium Ising model. In contrast to the conventional Ising model approach, this approach requires less computational costs and applies to data obtained from large-scale measurements. Further, it unveils the nonequilibrium nature of neuronal dynamics. We corroborate that the state-space kinetic Ising model captures the nonequilibrium statistics of the population activity by simulation studies, and demonstrate that it is applicable to large-scale parallel recordings using the Allen Brain Observatory data sets.
  • State-space analysis for neural population dynamics  [招待講演]
    Hideaki Shimazaki, Ken Ishihara, Ulises Rodriguez Dominguez, Sai Sumedh Hindupur, Miguel Aguilera, S. Amin Moosavi, Magalie Tatischeff, Jimmy Gaudreault, Christian Donner
    Neuro2022 2022年06月 口頭発表(一般) Okinawa 
    Co-variability of neural population activity reflects internal cognitive states of an animal, constrains information coding of sensory stimuli and behavior, and can provide insights into the mechanisms of underlying circuits. However, it is still challenging to elucidate the co-variability from spiking activities of neurons recorded from awake, behaving animals because they exhibit transient dynamics. Both individual firing rates and neurons’ co-activities can vary within experimental trials. To capture the non-stationary population dynamics, we extended a standard recurrent network model known as the Ising model/Boltzmann machine using a state-space framework. Here we introduce the current status and prospects of the research using the state-space Ising model. With the presented method, researchers can visualize the dynamics of the firing rates of individual neurons and their interactions from spiking data. It can also reveal macroscopic features of the neural system, including sparsity and interaction strength, and test whether the system is near a critical state, using thermodynamic quantities such as free energy, entropy, and heat capacity. All of these quantities are obtained in a time-resolved manner, making it possible to interpret them in comparison with the stimulus and behavioral paradigm of the experiments. We introduce our recent effort to include asymmetric kinetic dynamics in the state-space framework to account for the non-stationary and non-equilibrium neuronal dynamics. Here, measures of the non-equilibrium processes, such as conditional entropy and entropy production, can underpin the richness of population activity and the signature of the criticality appearing near non-equilibrium phase transitions. These data-driven methods offer valuable tools to characterize cortical states and their dynamical changes. They also elucidate population coding and underlying mechanisms when combined with appropriate encoding models. We introduce our Python-based analysis environment that automatically inspects basic spiking statistics, performs model selection, and visualizes time-varying neuronal interactions and macroscopic dynamics once users input their data.
  • 巽一厳, 松浦真人, 島崎秀昭, 稲村泰弘
    2021年度量子ビームサイエンスフェスタ
  • 標準リカレントネットワークモデルでつなぐ皮質回路の構造・機能・作動原理  [招待講演]
    島崎秀昭
    生理学研究所研究会「大脳皮質を中心とした神経回路:構造と機能、その作動原理」 2021年12月 口頭発表(招待・特別) 生理学研究所
     
    本講演では標準的なリカレントネットワークであるイジングモデル/ボルツマンマシンを用いて、皮質の回路構造と機能、そして作動原理を検証する研究の現状と将来展望を紹介する。イジングモデルはイベントによって相互作用する非線形素子からなるコンパクトなリカレントニューラルネットワークモデルで、統計学・統計物理学・機械学習における標準的なモデルとして広く使用されている。このモデルで神経活動を記述し解析することで、個々の要素に立脚した神経系の巨視的な挙動を明らかにできるだけでなく、神経系の計算を機械による計算と同じ枠組みで記述し、統一的に理解することが容易になる。本講演では、覚醒・行動下で記録された変動する神経活動を解析するために、我々が10年以上に渡って開発してきた状態空間法を用いたイジングモデルよる動的な集団活動の解析環境を紹介する。そして、これらの手法を用いたデータ解析に基づき、推定された活動の相関構造から皮質の回路構造を同定する研究を紹介する。次に、ネットワークの計算論的な機能を明らかにする研究として、神経細胞集団が有する外界の生成モデルをイジングモデルの枠組みで構成し、リカレントネットワークの神経応答のダイナミクスをベイズ推論における認識のダイナミクスとして記述する研究を紹介する。最後に、状態空間−イジングモデルによって神経細胞集団の外界への適応・学習のダイナミクスを可視化し、イジング型生成モデルの理論的な学習ダイナミクスと比較することで、脳の学習と認識の作動原理をデータから検証する取り組みを紹介する。
  • Bayesian Computation of Generic Neural Binary Code by Local Competition  [通常講演]
    Ulises Rodriguez Dominguez, Hideaki Shimazaki
    The 44th Annual Meeting of the Japan Neuroscience Society (Neuro 2021) 2021年07月 ポスター発表 
    The Bayesian brain hypothesis assumes that the neural systems seek causes of the observed sensory inputs using an internal model encoding how the environment generates the sensory inputs. The generative models of natural stimuli based on this hypothesis explained various features of early sensory systems, e.g., Gabor-like receptive fields of monkey or cat V1 neurons. The sparsity in the neural activity is one of the key constraints in successfully reproducing the neuronal features. However, these generative models assume that neurons encode stimulus by, potentially negative, continuous-valued firing rates. Although neurons communicate via spikes, the sparse generative model built on discrete spike events remains established. Here we propose a non-negative sparse spiking generative model, where sparsely-active, binary neurons encode external stimuli. By restricting the number of simultaneously active neurons, we impose sparsity in the population activity. In addition, local winner-take-all circuitry further promotes sparse local activity characterized by their uniquely structured higher-order interactions. We derived biologically-plausible learning rules following surprise minimization, in which we significantly reduced computational cost due to the limited number of active neurons. We demonstrate that the model successfully learns natural scenes with basis functions capturing spatial Gabor-like primitives, and that the sparsity improved the goodness-of-fit. These results promise the construction of more complex generative models based on spike events, which will allow us to test the Bayesian brain hypothesis using spiking activities of sensory cortices directly.
  • Miguel Aguilera, S. Amin Moosavi, Hideaki Shimazaki
    Entropy 2020: The Scientific Tool of the 21st Century 2021年05月 口頭発表(一般) Online 
    Many physical and biological dynamical systems operate away from thermodynamic equilibrium, driven by their own activity as well as their interaction with the environment. The kinetic Ising model is a prototypical model for studying such non-equilibrium dynamics. Since its behaviour is generally intractable for large sizes due to combinatorial explosion, mean field theories are often employed to approximate network dynamics. However, mean field methods are often unable to capture systems displaying long-range correlations such as those operating near critical phase transitions. To tackle this problem, different variants of mean field approximations have been proposed for kinetic Ising models, each making unique assumptions about the correlation structure of the system. This disparity complicates the challenge of systematically advancing beyond previous contributions. Here, using information geometry, we propose that existing methods can be described and extended in a unified framework. Our method is defined as a family of expansions (called Plefka expansions) of an intractable marginal probability distribution around a specific point of a simplified model, defined in an information geometric space. These points are obtained by an orthogonal projection to a sub-manifold of probability distributions displaying a simplified correlation structure. This approach not only unifies previous methods but allows us to define novel methods that make unusual assumptions for mean field methods, like models preserving specific correlations of the system. By comparing analytic approximations and exact numerical simulations in a kinetic Sherrington Kirkpatrick model, we show that the new approximations found by our method provide more accurate estimates of the dynamics of the systems than classical equations, even near critical phase transitions presenting large fluctuations. In sum, our framework unifies and extends existing mean field methods in the kinetic Ising model from an information theoretic view, constituting a powerful tool for studying the dynamics of complex systems.
  • 非定常・非平衡イジングモデルによる神経細胞集団活動の解明  [招待講演]
    島崎秀昭
    データ駆動生物学ワークショップ 2021年03月 口頭発表(招待・特別) 本田直樹
     
    本講演ではイジングモデルを用いた神経活動の解析の現状と将来展望を紹介する.脳の神経細胞はネットワークを形成し,スパイクと呼ばれるイベントを介して外界の情報や行動を符号化している.イジングモデルはイベントによって相互作用する要素からなるシステムを記述するコンパクトなモデルで,統計学・機械学習・統計物理学における標準的なモデルとして広く使用されている.このモデルで神経活動を記述することで,神経系の計算を機械や物理現象による計算と同じ枠組みで記述し,統一的に理解することが容易になる.神経科学における応用上重要なのは,データからモデルのパラメータを推定する「逆イジング問題」とその手法である.これまでに逆イジング問題を解くことで培養神経細胞や麻酔下の動物の集団活動が調べられてきた.しかし,これらは系の定常性と結合の対称性を仮定する平衡イジングモデルを用いており,覚醒・行動下で記録されたダイナミックに変動する神経活動を記述する事ができない.本講演では,我々が10年以上に渡って開発してきた状態空間法を用いたイジングモデルよる動的な集団活動の解析環境[1,2]を紹介し,最近の非平衡系への取り組み[3]を紹介する. [1] Shimazaki H, Amari S, Brown EN, and Gruen S (2012) PLOS Computational Biology 8(3): e1002385. https://doi.org/10.1371/journal.pcbi.1002385 [2] Donner C, Obermeyer K, Shimazaki H. (2017) PLoS Computational Biology 13(1): e1005309 https://doi.org/10.1371/journal.pcbi.1005309 [3] Aguilera M, Moosavi SA, Shimazaki H. (2021) Nature Communications. https://doi.org/10.1038/s41467-021-20890-5
  • Safura Rashid Shomali, S. Nader Rasuli, Hideaki Shimazaki
    The 3rd Sharif Neuroscience Symposium 2021年03月 
    We identify hidden neuronal motifs from the correlated activity of simultaneously recorded neurons. To elucidate hidden motifs of shared inputs to the observed neurons, we predict pairwise and triple-wise interactions of neurons under different motifs. Our tool is an analytical solution that accurately addresses the effect of strong/weak synaptic connections on spike-timing of leaky integrate-and-fire neurons receiving noisy inputs balanced near the neuron’s threshold [Shomali et al., 2018]. Comparing empirical interactions with predicted ones in the plane of triple-wise versus pairwise interactions, one can infer the hidden microcircuit that induces the observed interactions. We generalize this approach making it independent of the neuron models, by calculating the analytical boundaries for each motif, in extreme cases. Neurons in macaque V1 showed spare population activity characterized by positive pairwise but `negative’ triple-wise interactions [Ohiorhenuan et al., 2010]. A quantitative comparison reveals that the motif of excitatory-to-pairs explains the sparse activity, ruling out the shared inhibition [Shomali et.al, 2019]. We verify the robustness of this result to adaptation, mixing motifs, and to the condition in which the balanced inputs are slightly away from the threshold. Furthermore, we find evidence for this motif, in other brain regions in rodents. Sustained triplet correlations in OFC during predictable correct choices encodes behavioral decisions [Balaguer-Ballester et al, 2020]; we find most of the data show negative triple-wise and positive pairwise interactions for correct choices. Similarly, hippocampal CA3 neurons show negative 3rd-order, positive 4th-order, and negative 5th-order interactions [Shimazaki et al., 2015]. We extend our analysis to more than three neurons and present a table that links the strength and sign of higher-order interactions to hidden motifs. It affirms that such change of interactions’ signs is exclusively explained by excitatory-to-pairs. These imply that non-trivially, the excitatory-to-pairs motif is ubiquitous across regions and species.
  • 脳への計算論的アプローチ概説:視覚野の理論を中心に  [招待講演]
    島崎秀昭
    日本視覚学会2021年冬季大会 2021年01月 口頭発表(招待・特別) オンライン 日本視覚学会
     
    本講演では脳の計算論の発展の歴史を回路・情報・原理の3つの視点から紹介する.脳の回路と情報表現に対する理解は,高い変動性を示す神経スパイク活動を実現する回路・その制約下での情報処理,これら双方の解明を目指す過去20年の試みの中で深化してきた.その結果発見された興奮ー抑制の均衡ネットワークや我々の認識を制限する冗長な情報表現を紹介し,現在の大規模記録技術による検証を紹介する.脳の原理的記述には効率的符号化仮説・情報量最大化原理・スパース符号化等のさらに古い歴史があるが,現在では外界のモデル(生成モデル)が脳に構築されるとするベイズ脳仮説の枠組みによる記述に集約されつつある.近年この理論は,環境に対する働きかけを推論の枠組みに取り入れ,認識と行動の一体的理解を目指す「自由エネルギー原理」に拡張され,統一的な視点から回路・情報そして学習を見直す動きが広がっている.講演では現在に至る背景を紹介する.
  • 島崎秀昭
    CHAIN Webiner 2020年08月 公開講演,セミナー,チュートリアル,講習,講義等 オンライン 人間知・脳・AI研究教育センター
     
    本講演では,私たちが行ってきた神経科学の理論研究を紹介し,これから北大CHAINで行う研究を共有して皆さんと議論する機会としたい。神経細胞の複雑な集団活動を捉えるために,私たちは統計数理モデルを用いてその相互作用を可視化し,システムの特徴を捉える研究に取り組んできた。その結果明らかになる背後の回路構造や情報符号化を紹介する。加えてCHAINでは,生成モデルとしての脳の理解(ベイズ脳仮説)に取り組みたい。ベイズ脳仮説では神経の刺激応答活動は外界を推論するための事後分布を形成していると考え,これを支持する実験結果が蓄積されている。これを踏まえ,変分推論を実現する生物学的な神経回路網の数理的な理解を深めるとともに,数理モデルを介してデータから脳内の外界モデルを明らかにし,意識的体験に迫る研究を行いたい。特に初期感覚野では刺激応答活動の全てが感覚体験の報告に繋がるわけではなく,気づきや注意の効果はフィードバック回路を介した時間遅れ変調として現れる。この効果を定量化できる理論として提案しているニューラルエンジン理論を紹介する。従来のベイズ推論を熱力学的な見方で拡張することで得られるこの理論では,気づきや注意に関わる神経活動をエントロピーという指標で定量化できるだけでなく,生成モデルに基づきその内容にも踏み込むことができる。人間の意識的な体験の定量化にも適用可能と考えられ,CHAINにおける意識研究としてどのように位置付けられるか,哲学者・心理学者の皆さんも交えて議論したい。
  • Safura Rashid Shomali, Seyyed, Nader Rasuli, Hideaki Shimazaki
    Organization for Computational Neurosciences 2020 (CNS*2020) 2020年07月 
    Experimental studies demonstrated that neural populations exhibit correlated spiking activity that goes beyond pairwise correlations and involves higher-order interactions [1-5]. These higher-order interactions are known to encode stimulus information or the internal state of the brain [1-5]. However, the origin of this population activity and types of presynaptic neurons inducing the higher-order interactions remain unclear. Here we investigate how the interactions [6] among groups of 3, 4, and then N neurons emerge, when they receive common inputs on top of independent noisy background inputs, assuming simple connecting motifs. Given Poissonian common inputs, we calculate the neural interactions among clusters of neurons in a small time-window for the limit of the strong common input’s amplitude. When 2 or 3 neurons share excitatory/inhibitory common inputs, their pairwise and triple-wise interactions are well explained as functions of their baseline spontaneous rate, and the common-input’s rate [7]. We analytically solve the interactions for a cluster of more than 3 neurons when all of them share strong excitatory/inhibitory common input. Then, extending our analysis to the arbitrary number of N neurons we show that the N-th order interaction among neurons is still a simple function of the postsynaptic and common input rates. However, in larger populations, the N-th order interaction more strongly depends on the spontaneous rate of postsynaptic neuron rather than input rate. We also observe that larger number of neurons induce stronger magnitude of interactions, regardless of interaction’s sign. Moreover, shared excitatory inputs to all neurons always generate interactions with positive sign, while shared inhibitory inputs induce interactions with oscillatory signs with respect to N. Finally, we obtain the analytic result when excitatory or inhibitory inputs are shared among N-1 out of all N neurons: Surprisingly, the N-th order interactions exhibit signs opposite to those found when the common inputs is shared by all N neurons. In all mentioned cases, when the spontaneous activity of postsynaptic neurons is low, excitatory inputs can generate strong positive/negative higher-order interactions, whereas for high spontaneous activity, inhibitory neurons can induce large absolute values of higher-order interactions. These results are valid for any neuron model and solely based on the assumption of strong common inputs given to neurons! Since cortical, subcortical, and retinal neurons mostly exhibit spontaneous activity less than λ=40 Hz, for small time-window of Δ=5ms, these neurons are in low spontaneous regime i.e. λΔ < 0.2. Therefore we suggest that the significant higher-order interactions observed in retina, hippocampus, and cortices reveal that motifs of strong excitatory rather than inhibitory shared inputs are present and dominant there. Finally, we draw a table that links the strength of interactions and their signs to motifs, both for low and high spontaneous activity regimes. So based on interactions obtained from experimental data, it is possible to predict the underlying motif behind it. For example, for a specific experiment done in the hippocampal CA3 region [8], the observed negative 3rd-order, positive 4th-order, and negative 5th-order interactions leads us to the architecture of excitatory to pairs, that can generate such interactions simultaneously.
  • Effects of structured neural correlations in population coding: beneficial or detrimental?  [通常講演]
    Seyedamin Moosavi, Magalie Tatischeff, Bingyue Zhu, Hideaki Shimazaki
    Computational and Systems Neuroscience (Cosyne) 2020 2020年03月 ポスター発表 Denver, Colorado 
    In sensory cortices, information about external stimuli is conveyed by collective neural activities. Small neural populations often express perceptual sensitivity of psychophysical observers, suggesting that stimulus properties are redundantly coded by correlated neu- rons. This implies the existence of ‘information-limiting’ structures in noise correlations, which limit information as the population size increases. However, the existence of such correlations and their underlying mechanisms in cortical activities are yet largely unex- plored. Information conveyed by neural activities can be signified as the linear Fisher infor- mation that quantifies the sensitivity of population firing to stimulus changes. Here, from the viewpoint of information-geometry, we define detrimental correlations that decrease the linear Fisher information, and theoretically underpin the information-limiting correla- tions as a subset of the detrimental correlations. This generalizes the so-called differential correlations that were originally suggested as an exclusive form of the information-limiting correlations. We then show that detrimental correlations appear in activity of orientation- selective neurons in the monkey V1 area responding to drifting gratings [Kohn & Smith CRCNS.org]. Interestingly, observed correlations of small populations (∼10 neurons) en- hance sensitivity to the stimulus orientation, rather than decreasing its information. This positive effect was observed for covariance structures of both spike counts and binary spikes in fine temporal windows (∼10-100 ms), even after removing stimulus-driven co- variations caused by drifting of the sinusoidal gratings, using state-space models. This result implies that the correlations caused by the visual pathways act positively in coding the stimulus. Nevertheless, we find small structured correlations in coarse temporal-scales which have a detrimental role in population coding. The detrimental correlations appear as we increase the temporal window size of the analysis, indicating that they are caused by spatio-temporal correlations in longer time-scales. In summary, beneficial and detrimental correlations exist in respective time-scales of cortical activities, and the latter potentially limits the information in larger populations.
  • Inferring network motifs from neural activity using analytic input-output relation of LIF neurons  [通常講演]
    Safura Rashid Shomali, Majid Nili Ahmadabadi, Seyyed Nader Rasuli, Hideaki Shimazaki
    Computational and Systems Neuroscience (Cosyne) 2020 2020年02月 ポスター発表 Denver, Colorado 
    One of the challenges in Neuroscience is to identify network motifs in cortical microcircuitries from recorded neural activity. Having a nonlinear relation between synaptic inputs and output spikes helps us to narrow down possible architecture that can produce the observed activity. Nevertheless, theoretical approaches to this inverse problem remain challenging due to the lack of an analytic input-output relation of a neuron that operates under in vivo conditions. Here we present a method to uncover the network motifs from correlated population activity, using the recently proposed analytical solution of spike timing for a leaky integrate-and-fire neuron model that responds to strong synaptic inputs under noisy, balanced background inputs constituted by many weak synapses. Using this method, we found unique motifs behind the sparse population activity ubiquitously reported for neighboring cortical neurons, which is characterized by their positive pairwise and strong `negative’ triple-wise interactions. We show that two major motifs can express the sparse population activity: (i) inhibitory common inputs to three neurons and (ii) excitatory common inputs to pairs of neurons. By quantitatively comparing model prediction with the activity of macaque V1 neurons [Ohiorhenuan et. al. Nature 2010], we conclude that common inhibition cannot explain the observed sparse activity, but excitatory inputs to pairs well explain it. We quantitatively verify that when the spontaneous activity of postsynaptic neurons is low, as observed in the data, excitatory inputs to pairs can induce the strong negative interactions while inhibitory inputs to trios cannot. Accordingly, we also demonstrate that when some of the three neurons make directional or reciprocal connections, the motifs of excitatory inputs to pairs are required to induce the strong negative triple-wise interactions. These results indicate that local, excitatory inputs to neuron pairs constitute major cortical motifs that are involved in signal processing of in-vivo cortical microcircuitries.
  • Hideaki Shimazaki
    Computational Principles in Active Perception and Reinforcement Learning in the Brain 2020年02月 口頭発表(招待・特別) Kyoto University, Kyoto MACS International Symposium
     
    Neurophysiological studies on early visual cortices revealed that an initial feedforward-sweep of neural response depends on stimulus features whereas perceptual effect such as awareness and attention is represented as modulation of the late component (e.g., ~100 ms after the stimulus onset). The delayed modulation is presumably mediated by feedback connections from higher brain regions. Psychophysical experiments on humans using visual masking or transcranial magnetic stimulation showed that selective disruption of the late component vanishes conscious experiences of the stimulus. Here I provide a unified computational and statistical view on the modulation of sensory representation by internal dynamics in the brain, which provides a way to quantify the perceptual capacity of neural dynamics. A key computation is the gain modulation that represents integration of multiples signals by nonlinear devices (neurons). The gain modulation is ubiquitously observed in nervous systems as a mechanism to adapt neurons’ nonlinear response functions to stimulus distributions. It will be shown that the Bayesian view of the brain provides a statistical paradigm for the gain modulation. Moreover, the delayed gain-modulation of the stimulus response via recurrent feedback connections is modeled as a dynamic process of the Bayesian inference that combines the observation and top-down prior with time-delay. Interestingly, it will be shown that this process becomes a mathematical analogue of a heat engine in thermodynamics [1]. This view provides us to quantify the amount of the delayed gain modulation and its efficiency in terms of entropy changes of the neural activity. I will show how we can quantify the perceptual capacity from neural spiking data using the state-space Ising model of neural populations, which we have been developing in the past 10 years [2,3]. 1) Shimazaki (2015) Neurons as an Information-theoretic Engine. arXiv:1512.07855 (published as a book chapter) 2) Shimazaki, Amari, Brown, Gruen (2012) PLoS Comp Biol 8(3): e1002385 3) Donner, Obermeyer, Shimazaki (2017) PLoS Comp Biol 13(1): e1005309
  • Hideaki Shimazaki
    Combining Information theoretic Perspectives on Agency 2020年01月 口頭発表(一般) The University of Tokyo 
    Neurophysiological studies on early visual cortices revealed that an initial feedforward-sweep of neural response depends only on stimulus features whereas perceptual effect such as awareness and attention is represented as modulation of the late component (e.g., ~100 ms after the stimulus onset). The delayed modulation is presumably mediated by feedback connections from higher brain regions. Psychophysical experiments on humans using visual masking or transcranial magnetic stimulation showed that selective disruption of the late component vanishes conscious experiences of the stimulus. Here I provide a unified computational and statistical view on the modulation of sensory representation by internal dynamics in the brain, which provides a way to quantify the perceptual capacity of neural dynamics. A key computation is the gain modulation that represents integration of multiples signals by nonlinear devices (neurons). The gain modulation is ubiquitously observed in nervous systems as a mechanism to adapt nonlinear response functions to stimulus distributions. It will be shown that the Bayesian view of the brain provides a statistical paradigm for the gain modulation as a way to integrate an observed stimulus with prior knowledge. Furthermore, the delayed gain-modulation of the stimulus response via recurrent connections is modeled as a dynamic process of the Bayesian inference that combines the observation and prior with time-delay. Interestingly, it will be shown that this process is mathematically equivalent to a heat engine in thermodynamics. This view provides us to quantify the amount of the delayed gain modulation and its efficiency in terms of the entropy of neural activity. I will show how we can quantify the perceptual capacity from spike data using the state-space Ising model of neural populations, which we have been developing in the past 10 years. keywords thermodynamics, Bayesian brain, gain modulation, attention, neural engine
  • Hideaki Shimazaki
    シンギュラリティサロン 2019年12月 口頭発表(招待・特別) 東京 大手町サンケイプラザ シンギュラリティサロン
     
    本講演では脳の理論的理解を目指してきた計算論的神経科学・理論神経科学の発展の歴史を,それを支える主要な実験結果や他分野との関わりとともに紹介する.自然刺激への適応に基づく古典的な認識の理論から,外界のモデルを脳の中に持つとするベイズ脳仮説,そして環境に対する働きかけを推論の枠組みに取り入れ認識と行動の統一的理解を目指す自由エネルギー原理.これらがどのような実験事実に基づき(あるいは基づかず)構築され,情報理論・機械学習・統計物理といった他分野とどのような関わりのなかで発展してきたかを解説し,今後の展望を述べる.発展的試みのひとつとして動的にベイズ推論を実現する脳を情報論的なエンジンとして扱い,注意・意識的体験を含む脳の高次機能の説明を試みる新しいパラダイム「ニューラルエンジン仮説」を紹介したい.
  • Hideaki Shimazaki
    シンギュラリティサロン 2019年11月 口頭発表(招待・特別) グランフロント大阪・ナレッジサロン・プレゼンラウンジ シンギュラリティサロン
     
    本講演では脳の理論的理解を目指してきた計算論的神経科学・理論神経科学の発展の歴史を,それを支える主要な実験結果や他分野との関わりとともに紹介する.自然刺激への適応に基づく古典的な認識の理論から,外界のモデルを脳の中に持つとするベイズ脳仮説,そして環境に対する働きかけを推論の枠組みに取り入れ認識と行動の統一的理解を目指す自由エネルギー原理.これらがどのような実験事実に基づき(あるいは基づかず)構築され,情報理論・機械学習・統計物理といった他分野とどのような関わりのなかで発展してきたかを解説し,今後の展望を述べる.発展的試みのひとつとして動的にベイズ推論を実現する脳を情報論的なエンジンとして扱い,注意・意識的体験を含む脳の高次機能の説明を試みる新しいパラダイム「ニューラルエンジン仮説」を紹介したい.
  • Hideaki Shimazaki
    The 7th International Congress on Cognitive Neurodynamics 2019年09月 口頭発表(一般) Alghero, Italy 
    Stimulus information and cognitive states of an animal are represented by correlated population activity of neurons. The maximum entropy method provides a principled way to describe the correlated population activity using much less parameters than the number of possible activity patterns. This method successfully explained stationary spiking activity of neural populations such as in vitro retinal ganglion cells. Modeling activity of cortical circuitries in vivo, however, has been challenging because both the spike-rates and interactions among neurons can change according to sensory stimulation, behavior, or an internal state of the brain. To capture the non-stationary interactions among neurons, we augmented the maximum entropy model (Ising model) using a state-space modeling framework, which we call the state-space Ising model. We will demonstrate that applications of the state-space Ising model to activity of cortical neurons reveal dynamic neural interactions, and how they contribute to sparseness and fluctuation of the population activity as well as stimulus coding.
  • Judging between Excitation and Inhibition: Identifying Local Network Architecture by an Analytic Pre-Post Relation  [通常講演]
    Safura Rashid Shomali, Majid Nili Ahmadabadi, Seyyed Nader Rasuli, Hideaki Shimazaki
    Bernstein Conference 2019 2019年09月 ポスター発表 Berlin 
    Recognizing network architecture exploiting the recorded activity of neurons only, is a challenge in neuroscience. In general, this is a hard task; but an analytic tool can help us to narrow down possible scenarios and approach the architecture behind the activity. Recently, researchers have found a statistical pre-post relation for a Leaky-Integrate-and-Fire (LIF) neuron when neuron receives signaling input on top of noisy background inputs near the threshold regime [1]. We use this analytic relation in mixture models to investigate the effect of shared inputs on network architecture. It connects synaptic inputs and statistics of population activity (pairwise and higher-order correlations) to network architecture in basic symmetric and asymmetric motifs; this lets us identify the underlying network architecture. We use it to address the architecture behind sparse population activity, reported for monkey's V1 neurons [2]. Comparing the theoretical graphs with the experimental data [2,3], we determine whether the underlying architecture is symmetric or asymmetric, whether synapses are excitatory or inhibitory, and more. We also consider the possibility of recurrent activities among neurons; which we categorize in 16 main motifs. The study on shared inputs and recurrent connections shows the main structure in which one excitatory common input gives synapses to a pair of neurons is responsible for the observed strong negative triple-wise correlations. This is in contrast with the intuitive expectation that shared inhibition causes the observed sparse activity. Finally, we ask why excitatory to pairs, and not inhibitory based architectures, induce such strong correlations in the aforementioned experiment. We attribute it to the sparseness of neuronal activity: When neurons are predominantly silent; an excitatory input causes a more significant change compared to any inhibitory one.
  • Miguel Aguilera, Amin Seyed Moosavi, Hideaki Shimazaki
    15th Granada Seminar: Stochastic and Collective Effects in Neural Systems 2019年09月 ポスター発表 Spain 
    Advances in high-throughput data acquisition technologies provide unprecedented possibilities for the statistical description of very large biological systems. In this scenario, there is a pressing demand for mathematical tools that can cope with the large data sets. Asymmetric kinetic Ising models are powerful tools for studying dynamics of the complex systems, as well as for analyzing experimental recordings using parameters inferred from the data. However the models’ behaviour is not analytically tractable for large networks. Therefore the mean field theories are often employed to obtain its approximate statistical properties. Many variants of the classical naive and TAP (i.e., the second-order) mean field approximations have been proposed, each of which makes unique assumptions on correlation structure among its elements (e.g., see [1,2]). A jumble of the methods makes it obscure how one can systematically advances these previous achievements. Here, we propose a unified framework of the mean field theories for the dynamics of kinetic Ising systems from the information geometric viewpoint. The framework builds on Plefka expansion [3,4] of the model around a specific point projected on a sub-manifold of probability distributions that represent the assumed correlation structure. This approach not only unifies the previous methods but also can propose novel mean field approximations, and is applicable to approximate stationary and transient behaviour of the system. By comparing the analytic approximations with exact numerical simulations, we demonstrate that the newly proposed approximations more accurately estimates evolution of equal-time and time-delayed covariance structure than the classical ones. We also demonstrate that estimation of the parameters using the proposed approximations in inverse Ising problems outperform the others. Our framework organizes various attempts to the mean field approximations of kinetic Ising models, and naturally brings about better approximation methods.
  • 学習と認識の熱力学:ニューラルエンジンとはなにか?  [通常講演]
    島崎 秀昭
    生理研研究会2019 認知神経科学の先端 「脳の理論から身体・世界へ」 2019年09月 口頭発表(招待・特別) 岡崎 生理学研究所
     
    本講演では脳が外界を学習し,推論する仕組みを情報理論・機械学習・ベイズ統計学の視点から概観し,加えて神経活動のエントロピー変化に基づいて脳の学習・推論を記述する熱力学的な枠組みを紹介する.脳の学習・推論は順方向結合・リカレント結合からなる神経回路網の非線形なダイナミクスによって実現される.感覚野では順方向結合による初期応答は刺激特徴のみに依存し,気づき・注意・報酬価値等の有無にかかわる内発的な状態は初期応答後の活動に対する変調として表れることが報告されている.この遅延変調は高次領野からのリカレント結合に起因するものと考えられている.逆向マスキングやTMS刺激を用いた人に対する研究では,この遅延部分の活動の選択的阻害により意識的体験が消失することが示されている.講演では刺激応答の神経ダイナミクスが観測と事前知識を統合する動的なベイズ推論として記述されることを示し,さらにリカレント回路による遅延変調を伴う推論過程が情報論的なエンジンを構成することを示す(ニューラルエンジン).これにより感覚入力に対する内発的な変調の程度やその効率を神経活動のエントロピー変化に基づいて定量化する方法を提案する.
  • Hideaki Shimazaki
    Data Science, Statistics, & Visualisation (DSSV2019) 2019年08月 口頭発表(招待・特別) Kyoto, Japan International Association for Statistical Computing
     
    Neurons in the brain communicate each other using electrical pulses known as spikes. Information of external world in the brain is therefore represented in spiking activity of neuronal networks. However it is difficult to decipher how they code the information based on direct inspection of the spike sequences due to their excessive variability, which is not merely noise but a hallmark of cortical information processing realized by balanced activity of excitatory and inhibitory neurons. Thus we need visualization techniques that uncover statistical structure of the neural population activity. Using binary variables to represent presence or absence of spikes in a small time window, one can describe synchronous/asynchronous activity of multiple neurons as sequences of binary patterns. By fitting an undirected graphical model to such data, one can visualize statistical network structure of the simultaneous activity of neurons. However this conventional model is limited because it assumes static neural activity (i.e., constant spike rates and correlations), which could be realized only in in-vitro networks or neurons under anesthetized conditions. In-vivo neurons recorded from awake animals engaged in tasks is more dynamic: The activity rates of neurons dynamically changes in response to stimulus presentation, and strength of correlations between neurons may change. These statistical features are not only regulated by the stimulus but also an internal state of the brain. To account for such dynamics, we introduce a state-space model of neural population activity, and visualize coordination of the neural activity using a time-varying graph \cite{shimazaki2012state,donner2017approximate,gaudreault2018state}. In this talk, I will explain how this visualization technique reveals macroscopic properties of the network dynamics, and how it contributes to neural coding studies.
  • Jimmy Gaudreault, Arunabh Saxena, Hideaki Shimazaki
    The 2019 International Joint Conference on Neural Networks (IJCNN) 2019年07月 口頭発表(一般) Budapest INNS, IEEE CIS
     
    Sequences of correlated binary patterns can represent many time-series data including text, movies, and biological signals. These patterns may be described by weighted combinations of a few dominant structures that underpin specific interactions among the binary elements. To extract the dominant correlation structures and their contributions to generating data in a time-dependent manner, we model the dynamics of binary patterns using the state-space model of an Ising-type network that is composed of multiple undirected graphs. We provide a sequential Bayes algorithm to estimate the dynamics of weights on the graphs while gaining the graph structures online. This model can uncover overlapping graphs underlying the data better than a traditional orthogonal decomposition method, and outperforms an original time-dependent Ising model. We assess the performance of the method by simulated data, and demonstrate that spontaneous activity of cultured hippocampal neurons is represented by dynamics of multiple graphs.
  • 神経細胞集団活動の統計解析:脳の熱力学に向けて  [通常講演]
    島崎 秀昭
    サロン・ド・脳 2019年06月 京都 サロン・ド・脳運営委員会
     
    脳の神経細胞はスパイクを介し協調して情報を処理している.相互作用する神経細胞集団に対し,講演者はこれまで統計学・機械学習・情報幾何・統計物理学・熱力学の知見を横断的に使用して,その情報処理に迫る手法を開発してきた.本講演の前半では単一神経細胞の発火頻度推定・複数神経細胞活動の相関構造の推定手法を解説し,これにより明らかになる神経活動の高次統計構造・背後のネットワーク構造・刺激情報源の符号化を紹介する.特に多体の相互作用を記述する統計物理モデルを時系列解析へ拡張して覚醒動物の神経活動に適用することで,個々の神経細胞の記述に基づいて系の巨視的な量(熱力学的量)のダイナミクスを推定できることを示す.後半では脳のダイナミクス・学習に対する新しい見方を提案する.脳の統一理論として推論・学習・行動を変分自由エネルギー最小化で説明する理論(自由エネルギー原理)が提案され,従来の予測符号化理論・ベイズ脳仮説を包含した形で研究が進行している.ここではこの理論における推論のダイナミクス・生成モデルの学習に対して熱力学的な取り扱いを導入し,推論(神経応答)のダイナミクスに対するエントロピー保存則(第一法則)を導き,学習がエントロピー増大則(第二法則)として表される条件を示す [1].ベイズ推論を実現する神経ダイナミクスでは,初期の刺激応答に対しフィードバック回路による事前知識が時間遅れで融合し,事後分布としての神経応答が形成される.第一法則に基づき,この時間遅れの変調を伴う神経ダイナミクスが熱機関と同様に取り扱えることを示す(ニューラルエンジン仮説)[2].多くの電気生理実験で刺激に対する初期応答ではなく遅い応答コンポーネントの変調にコンテクスト・注意・気づき・報酬価値等の情報が表されると報告している.脳を情報論的なエンジンと見ることで,能動的な刺激変調を定量化しその効率を計算できる.脳の熱力学的な解析に向けて理論と検証のツールが整ってきたことを示したい. 参考文献 [1] Shimazaki H. The principles of adaptation in organisms and machines I: machine learning, information theory, and thermodynamics. (2019) arXiv:1902.11233 [2] Shimazaki H. Neurons as an Information-theoretic Engine. (2015) arXiv:1512.07855 (Book chapter, Springer 2018)
  • Hideaki Shimazaki
    2019年03月 口頭発表(一般) OIST Kenji Doya
     
    Population activity of neurons is constrained by their biophysical properties and network architecture. Hence, statistical regularity of neural activity in turn tells us about these underlying mechanisms. In this talk, we report structured higher-order interactions in population activity of hippocampal CA3 neurons in cultured slices, and show that it is explained by excess simultaneous silence of neurons [Shimazaki et al. Sci Rep 2015]. We then investigated mechanisms underlying the excess silence. Using an analytical input-output relation of a leaky integrate-and-fire neuron model that operates under in-vivo conditions [Shomali et al. J Comp Neuro 2017], we constructed a model that links synaptic inputs, nonlinearity of spiking, network architecture, and statistics of population activity. With this model, we found that either common inhibition of neurons or excitatory inputs to pairs of neurons could induce the excess silence; but only the latter quantitatively explained the observed level of excess silence [Shomali et al. biRxiv 2018]. This work shows that the unified modeling framework is a useful tool to explore biophysical mechanisms of neural information processing from population activity.
  • Magalie Tatischeff, Jimmy Gaudreault, Christian Donner, Hideaki Shimazaki
    Consciousness Research Network (CoRN 2019) 2019年02月 ポスター発表 Okazaki, Japan
  • S. Amin Moosavi, Hideaki Shimazaki
    Consciousness Research Network (CoRN 2019) 2019年02月 ポスター発表 Okazaki, Japan
  • 社会神経科学的アプローチによる精神疾患の社会性障害の理解  [招待講演]
    島崎秀昭
    平成30年度生理研研究会 第8回社会神経科学研究会 2018年11月 シンポジウム・ワークショップパネル(指名) 自然科学研究機構 生理学研究所(愛知県岡崎市) 高橋宗良(玉川大学 特任准教授)・定藤規弘(生理研 教授)
  • Hideaki Shimazaki, Magalie Tatischeff, Jimmy Gaudreault, Christian Donner
    Analysis and Synthesis for Human/Artificial Cognition and Behaviour 2018年10月 ポスター発表 
    Methods in thermodynamics and statistical mechanics are useful to understand activity of interacting neurons that collectively process information. Most approaches that use a model in statistical mechanics, namely an Ising model, assume stationary data. However, the correlated activity may organize dynamically during behavior and cognition. This poster presents (i) our recursive Bayesian estimation methods that can optimally estimate the time-varying interactions among neurons (Shimazaki et al. IEEE ICASSP 2009, PLOS CB 2012, JPCS 2013), (ii) the results of application to various data sets, including dynamics of thermodynamic quantities of the networks in relation to task paradigms (Donner et al. et al. PLOS CB 2017; Gaudreault et al. ICANN 2018), and (iii) our 'Neural Engine' hypothesis on neural dynamics for efficient encoding of stimulus information, which claims that neurons act similarly to a heat engine when its response to sensory stimulus is enhanced by gain-modulation by delayed feedback mechanisms (Shimazaki arXiv 2015; Springer 2018).
  • Jimmy Gaudreault, Hideaki Shimazaki
    The 27th International Conference on Artificial Neural Networks (ICANN2018) 2018年10月 口頭発表(一般) Rhodes, Greece European Neural Network Society
  • The active inference in decision making by adult zebrafish revealed by in-vivo imaging of the telencephalic neural activities in the closed-loop virtual reality environment  [通常講演]
    Makio Torigoe, Islam Tanvir, Hisaya Kakinuma, Fung Chi Chung Alan, Takuya Isomura, Hideaki Shimazaki, Tazu Aoki, Taro Toyoizumi, Tomoki Fukai, Hitoshi Okamoto
    The 41st Annual Meeting of the Japan Neuroscience Society 2018年07月 ポスター発表 
    Selecting a logical behavioral choice from the available options, i.e. decision making is essential for animals. Recently, adult zebrafish has drawn attention as a model animal for the study of decision making due to its capability of various adaptive behaviors and the conservation of the basic telencephalic structure throughout the evolution of the vertebrates, which includes the evolutionary functional homologs of the hippocampus, amygdala, cortex, and basal ganglia. However, how the decision making is made by the interaction of neurons of these multiple areas in the telencephalon has largely remained unknown. In the present study, we aimed at directly addressing this process by establishing the closed-loop virtual reality (VR) system for the head-tethered adult zebrafish with the 2-photon calcium imaging system. Besides conservation of its structure in comparison with other vertebrates, the small size of the zebrafish telencephalon allows us to capture the neural activities from wide and multiple regions at the cellular levels. The adult zebrafish harboring G-CaMP7 in the excitatory neurons were trained to perform visual-based active and passive avoidance tasks. Furthermore, after learning was once established in the closed-loop condition, we suddenly removed the visual feed-back to make the system open-loop. The Non-negative Matrix Factorization (NMF) Analysis revealed the one ensemble of neurons whose activities were suppressed by the recognized backward movement of the landscape, and the other ensemble suppressed by reaching the goal compartment. These ensembles recovered throughout the trials under the open-loop condition. These results suggest that these two ensembles encode the prediction errors between the status represented by the real sensory inputs and the favorable status to achieve to successfully escape from the danger, i.e. visual inputs of the backward moving landscape and the wall color of the goal compartment, and the behaviors are taken so that these errors become minimum. Our result supports that the adult zebrafish behaves in decision making based on the active inference in the free energy principle, where agents take actions to suppress the prediction errors by trying to make the internal representation of the bottom-up sensory states match those of the top-down predictions, and demonstrate the strong conservation of the basic principle of decision making throughout the evolution.
  • 島崎 秀昭
    第6回 数理モデリング研究会 2018年07月 公開講演,セミナー,チュートリアル,講習,講義等 軽井沢 
    神経細胞の集団活動のメカニズムについて考察した最近の研究を紹介する.神経細胞集団の発火活動の高次統計構造を調べるとスパースな発火活動を示していることが知られている [1,2] がその生成メカニズムは明らかになっていない.今回,生体内の条件に近い状態における積分発火ニューロンの解析的な入出力関係を導出した[3]のを機会として,この解を用いて共通入力によって生成される集団発火活動の相関構造を調べた.スパースな発火を生成する複数の神経回路を調べて視覚野神経細胞の活動と比べることで,基本となる回路網構造の候補を絞り出したので紹介する.
  • State-space analysis of an Ising model reveals contributions of pairwise interactions to sparseness, fluctuation, and stimulus coding of monkey V1 neurons  [通常講演]
    Jimmy Gaudreault, Arunabh Saxena, Hideaki Shimazaki
    脳と心のメカニズム第18回冬のワークショップ 2018年01月 ポスター発表
  • HIDEAKI SHIMAZAKI
    Fluctuations of event occurrences in a variety of networks 2017年11月 口頭発表(一般) Kyoto University Shigeru Shinomoto (Kyoto University), Takaaki Aoki (Kagawa University) Ryota Kobayashi (NII), Taro Takaguchi (NICT), Hideaki Shimazaki (Kyoto University, HRI)
     
    In this talk, we will summarizes reported findings on variable and sparse population activity, and suggest computational models that explain it. Analysis of higher-order interactions among spiking activities of neurons revealed sparse population activity, which were explained by several computational models of neurons with different network architectures. We compared prediction by the computational models with neural interactions observed in monkey V1 neurons, and suggest a common neural network structure underlying visual processing.
  • 神経細胞集団活動の統計数理  [招待講演]
    島崎 秀昭
    玉川大学 脳科学ワークショップ 2017年09月 口頭発表(招待・特別) 山梨県笛吹市 玉川大学
     
    脳の神経細胞はスパイクと呼ばれるイベントを介して情報のやりとりを行い,協調して情報を処理している.このように相互作用するイベント生成システムの記述として,講演者はこれまで統計学・機械学習・統計物理学・熱力学の知見を横断的に使用して神経細胞の情報処理にせまる手法を開発してきた.本講演ではこれらの手法により明らかになった神経細胞の集団活動の特徴を紹介する.
  • 鳥越 万紀夫, Islam Tanvir, 柿沼 久哉, 島崎 秀昭, 馮 志聰, 青木 田鶴, 深井 朋樹, 岡本 仁
    第40回日本神経科学大会 2017年07月 口頭発表(一般) 幕張メッセ 日本神経科学会
     
    One of major goals in neuroscience is to understand how the dynamics of neural activities across multiple brain regions generate function. For this, model animal, in which we can simultaneously capture the neural activity from multiple brain regions, is necessary. In our project, we focused on adult zebrafish as a model animal which has conserved brain regions to mammal and ability to learn and can be used for examining the neural mechanisms for decision making. Selecting a logical behavioral choice from the available options, i.e. decision making is essential for animals. Accumulating evidence implicates the contribution of cortico-basal ganglia circuits, the hippocampus and amygdala in decision making. The hippocampus and amygdala encode external information and its value, respectively. The cortico-basal ganglia circuits affected by these two regions make and execute behavior programs along with learning, implying the dynamic and coordinated activity patterns in these three regions. In the present study, we aimed at directly capturing this process. To achieve this aim, we established the closed-loop virtual reality (VR) for tethered adult zebrafish with 2-photon microscopy imaging system. In this system, adult zebrafish performed visual-based active and passive avoidance task and simultaneously their telencephalic neural activities containing putative three regions was imaged. Transgenic fish harboring G-CaMP7 in the excitatory neurons in the pallium (camk2a:GVP; UAS:G-CaMP7; nacre or casper) were used. The fish could learn the VR task and the neural activities during learning were captured at the cellular level. Only in the success trials, specific neural activity patterns and increment of recruited neural ensembles were observed, suggesting that, for proper decision making, specific neural ensembles were recruited. Additional analysis revealed neural ensembles coding visual, movement and learning information. These results provide crucial clues to understand the neural mechanisms of decision making.
  • 神経活動データの時系列モデリング入門  [招待講演]
    島崎 秀昭
    第8回脳科学若手の会合宿 2017年03月 口頭発表(招待・特別) 脳科学若手の会
     
    脳の神経細胞ネットワークはスパイクを伝達して情報処理を行っています.私たちの外界に対する認識や行動はこの離散的信号に書き込まれていると言ってよいかもしれません.従ってこの時系列信号を扱う統計・数理が必要です.スパイク活動データの統計構造は外界からの入力に依存するだけでなく,ネットワーク構造やスパイク生成・シナプス入力にまつわる非線形性によっても規定されます.このことは統計モデルを用いて神経活動を調べることで,背後のネットワークや非線形演算を担うメカニズムに関する知見が得られることを意味します.またスパイク時系列の統計モデルは情報を担う特徴量を検証する神経コーディング研究に欠かすことのできないツールであり,その応用として神経活動から人・動物の行動や意図をデコーディングする神経補綴技術があります.本ワークショップでは多細胞同時記録スパイク時系列の代表的な解析法を紹介し,実験を主とする学生・ポスドクを対象に統計モデリングに親しんでもらうことを目的とします.イベント時系列を扱う点過程モデリングの基礎をポアソン過程から始めて丁寧に説明し,応用として一般化線形モデルによるネットワークモデリング・状態空間モデルによる動物の行動の逐次推定技術を学びます.またボルツマン機械と呼ばれる機械学習の手法を紹介し,集団活動パターンの解析を行います.このモデルは相互作用する要素からなるシステムを記述するコンパクトなモデルで,統計学・機械学習・統計物理・熱力学の交差点に位置し,生物と機械の計算を統一的に記述することができます.この観点から神経回路網活動のダイナミクスの熱力学的な解析についても触れたいと思います. キーワード 点過程モデル・一般化線形モデル・指数分布族・イジングモデル・ガウス過程・最大エントロピーモデル・カルマンフィルタ・ニューラルエンジン
  • Exact analysis of spike­timing and higher­order interactions of neurons at the threshold regime suggests network architecture underlying sparse population activity  [通常講演]
    S. R. SHOMALI, M. NILI AHMADABADI, S. RASULI, H. SHIMAZAKI
    Society for Neuroscience 2016 2016年11月 ポスター発表 
    The accurate prediction of how spiking of a presynaptic neuron affects spike timing of a postsynaptic neuron in vivo has significant importance in a variety of questions in Neuroscience. An exact solution for this problem under conditions resembling in vivo, however, is lacking due to the nonlinearity of the neuron’s spike generation mechanism. Neural activity in vivo exhibits significant variability. It is suggested that this variability reflects neuronal activity near the threshold regime, where small fluctuations of presynaptic neurons can significantly affect postsynaptic spike­timing. Here, we analytically investigate impact of a signaling input on a leaky integrate­and­fire neuron that receives background noise at the threshold regime. The signaling input models a synaptic or an assembly of nearly synchronous synaptic activity that conveys information while the background noise represents ongoing activity of many weak synapses. We demonstrate that, at the threshold regime, it is possible to obtain an exact solution that explains how the signaling input changes the spike­ timing distribution. We then use the result to predict higher­order interactions of population activity caused by shared signaling inputs under different network architectures. This prediction allows us to uncover network architecture behind the population activity. In particular, we suggest architecture behind sparse population activity observed in monkey V1 or rat hippocampal neurons, which involves higher­order interactions. Contrary to common intuition, we find it unlikely that common inhibition causes the sparse activity; instead, we quantitatively show that the observed activity can result from local excitatory common inputs by comparing the theoretical prediction with empirical results obtained from the monkey V1neurons.
  • Estimating dynamic functional networks of larger neural populations  [通常講演]
    C. DONNER, H. SHIMAZAKI
    Society for Neuroscience 2016 2016年11月 ポスター発表 
    Information in the brain is encoded in spiking activity of neural populations. However the number of activity patterns that a population can generate increases exponentially with the number of neurons. The maximum entropy method provides a principled way to describe the population activity using a tractable amount of parameters. This method successfully explained stationary spiking activity of neural populations by using less features, such as spike-rates of individual neurons and interactions among pairs of neurons, than the number of possible activity patterns. Modeling activity of cortical circuitries in vivo, however, has been challenging because both, the spike-rates and interactions, can change according to sensory stimulation, behavior, or an internal state of the brain. To capture the non-stationary functional network activity, a state-space framework was suggested to model dynamics of the neural interactions (Shimazaki et al. PLOS Comp Biol, 2012). However, based on exact analysis, the method suffers from computational cost; therefore its application was limited to only ~15 neurons. Here we introduce multiple analytic approximation methods to the state-space model, and make it possible to estimate dynamic pairwise interactions of up to 30 neurons. More specifically, we applied the pseudolikelihood approximation to the neural interaction model, and combined it with the TAP mean-field or Bethe approximation methods to carry out sequential Bayesian estimation of the model parameters. We found that belief propagation algorithm finds the solution of the Bethe approximation fast, and the provided solution is more precise than the TAP method. However, the belief propagation method sometimes fails to converge to a unique solution. Here we propose a hybrid method in which we use an alternative approach, the concave convex procedure (CCCP) that guarantees convergence, when a solution was not found by belief propagation. We compare the performance of these methods using simulated data, and demonstrate applicability of the method to experimental data recorded from awake animals. In addition to the time-varying interactions, the method allows us to investigate dynamics of global properties of recorded networks, such as entropy or sparsity of the population activity.
  • Large-scale inference of time-varying neural interactions  [通常講演]
    Christian Donner, Hideaki Shimazaki
    ICONIP2016 2016年10月
  • HaDi MaBouDi, Hideaki Shimazaki, Lars Chittka
    EURBEE 2016 2016年09月 
    The honeybee olfactory system is a well-established model for understanding functional mechanism of learning and memory. Olfactory stimuli are first processed in the antennal lobe (AL), and then transferred to the mushroom body (MB) and lateral horn (LH) through dual pathways termed medial and lateral antennal lobe tracts (m-ALT and l-ALT). Recent studies reported honeybees could perform elemental learning by associating an odour with a reward signal even after lesions in m-ALT or blocking MB [1,2] although such learning has been studied by modelling m-ALT that terminates on MB. To test if the lateral pathway (l-ALT) is sufficient for elemental learning, we modelled local computation within glomeruli with axons of projection neurons (PNs) connecting to a decision neuron (LHN) in LH. The model is further enriched by synaptic plasticity in AL and octopaminergic modulation in AL and LH [4-6]. We show that inhibitory spike-timing dependent plasticity (iSTDP) on synapses from local neurons (LNs) to PNs, a model of non-associative plasticity by exposure to different stimuli, decorrelates PNs’ outputs. By additionally modelling octopaminergic effects on synapses among LNs in AL and PNs into LHN using modulated STDP [3,7], the model can discriminate conditioned stimuli, which explains associative olfactory learning by a few stages of odour-processing in the l-ALT. Importantly, by monitoring performance of models with different connectivity caused by non-associative learning, one can describe changes in structural organization of honeybees' AL [8] and their behavioural performances during the first week of their life.
  • Analytical study of correlation and Fisher information caused by common inputs  [通常講演]
    Safura Rashid Shomali, Majid Nili Ahmadabadi, Hideaki Shimazaki, S Nader Rasuli
    the 39th Annual Meeting of the Japan Neuroscience Society 2016年07月 ポスター発表 
    One of the fundamental questions in neuroscience is the way information is processed through neurons in cortex. Neural correlation is one of the key features of neural activity that gives a clue to this question; because it reflects underlying network architecture and their computation. However the relation between correlation and information, that the neurons convey, has not been described with full mathematical clarity even for simple neuron models and network architectures. To tackle this problem, here we consider one-layer feedforward leaky integrate-and-fire (LIF) neurons where each receives noise and common inputs. For this network we are able to analytically study pairwise and higher-order correlations caused by the common inputs. Based on our previous study, we found analytically the exact spike-timing distribution of the LIF neurons when they receive signaling input on top of noisy balanced inputs. This spike-timing distribution is used not only to compute the neural correlation but also to analytically calculate Fisher information with respect to the common input amplitude. We elucidate how pairwise, higher-order correlation and Fisher information are changed as a function of the noisy input’s diffusion coefficient and common input amplitude. Based on these calculations, we discuss how the Fisher information and correlation caused by common input’s amplitude are related in the space of scaled diffusion coefficient and input amplitude’s parameters. We observe an optimum value for the common input amplitude, in which, the Fisher information is maximized. Finally, we investigate if this optimum value coincides with the so called equilibrium value which is obtained, using spike timing dependent learning procedures.
  • Exact spike-timing distribution and its usage in neural structure identification  [通常講演]
    S. Rashid Shomali, M. Nili Ahmadabadi, H. Shimazaki, S.N. Rasuli
    Statistical physics methods in biology and computer science (Satellite meeting of StatPhys 2016 2016年07月 ポスター発表 Ecole Normale Superieure, Paris 
    The accurate prediction of how spiking of a pre-synaptic neuron affects spike timing of the post-synaptic neuron has significant importance in a variety of questions of neurosciences. It is crucial in understanding information transfer, timing dependent learning, reverse engineering of the network architecture and more. Such prediction, however, is lacking due to the nonlinear nature of neuron's spiking dynamics; the neuron exclusively fires if its membrane potential reaches the so-called threshold potential. The Fokker-Planck equation, which is used to find such accurate prediction, is hardly tractable in the presence of absorbing boundary condition, imposed by the very threshold criteria. However, there is a specific but ubiquitous situation for which we can solve the Fokker-Planck equation; it happens if the mean current which a neuron receives equals threshold potential*. We have been able to exactly solve the aforementioned problem of prediction for this threshold regime. Then we used our exact solution to reverse engineer the network architecture of monkey's cortex, in V1, using existing experimental data. It is notable that the architecture we have found contradicts with the intuitive structure which is widely accepted.
  • Toward thermodynamic principles of consciousness  [通常講演]
    島崎 秀昭
    Seminar 2016年07月 口頭発表(一般) Sussex University Christopher L Buckley
     
    Population activity of neurons is constrained by their biophysical properties and network architectures. Thus neurons are statistically dependent each other, and collectively process information. Methods in thermodynamics and statistical mechanics are particularly useful to understand such collective activity of interacting elements. Previous approaches that use models of statistical mechanics assumed stationary neural data. However, it could be that the correlated activity organizes dynamically during behavior and cognition, and this may be independent from spike rates of individual neurons. To resolve this issue, we have been developing a recursive Bayesian estimation method that optimally estimates time-varying neural interactions [1]. I will introduce a result of dynamic (higher-order) interaction of neurons in monkey M1 that is correlated with monkey’s intention to move an arm [1], as well as recent findings on structured higher-order interactions of hippocampal neurons and possible underlying mechanisms [2]. In addition, recent development in approximate methods for large-scale analysis of neural populations will be discussed. The large-scale analysis allows us to characterize dynamic thermodynamic quantities of networks such as entropy, sparsity, and sensitivity. These methods are being developed in part to test ‘Neural Engine Hypothesis’ proposed in [3], my theory toward ‘consciousness’ based on thermodynamic principles. I will demonstrate that neural systems act similarly to a heat engine when an organism actively modulates incoming sensory information as is expected for e.g., attention or reward modulation. This view allows us to quantify the amount of internal computation in a brain, and suggests that an organism is `conscious’ when it acts as an information-theoretic engine. [1] Shimazaki H, Amari S, Brown EN, and Gruen S. State-space analysis of time-varying higher-order spike correlation for multiple neural spike train data. PLOS Comput Biol (2012) 8(3): e1002385 [2] Shimazaki H, Sadeghi K, Ishikawa T, Ikegaya Y, Toyoizumi T. Simultaneous silence organizes structured higher-order interactions in neural populations. Sci Rep (2015) 5, 9821 [3] Shimazaki H. Neurons as an Information-theoretic Engine. arXiv:1512.07855 (2015)
  • Higher-order interactions of neural populations  [通常講演]
    島崎 秀昭
    Seminar at TNJC 2016年07月 口頭発表(一般) UCL Gatsby Mehdi Keramati
     
    Population activity of neurons is constrained by their biophysical properties and network architectures. By accurately assessing its statistical structure, and comparing it with model prediction, we gain insights into the underlying networks and their coding principles. The maximum entropy (MaxEnt) principle has been successfully used to describe the population activity from limited amount of data. Conventionally, the MaxEnt model is characterized by parameters for interactions among subset neurons. Recent studies show that neural populations express significant higher-order interactions (HOIs), namely interactions among more-than 2 neurons. However, the previous studies have not identified a key feature that can summarize the seemingly diverse HOIs. Here we examined HOIs of the hippocampal CA3 neurons in cultured slices [Shimazaki et al. Sci Rep 2015], and report that most neurons that expressed HOIs exhibited significantly longer periods of simultaneous silence than predicted by a pairwise MaxEnt model. This simultaneous silence explained ~20% of the entropy related to the HOIs. We also confirmed presence of the structured HOIs predicted from the simultaneous silence. Through two different modeling approaches, we demonstrate that population activity caused by locally shared inputs and nonlinear thresholding reproduces the structured HOIs, and that this structure conveys information of input. This ubiquitous structure of HOIs observed in the activities of both experimental and model neural populations suggests that neurons are operating in a unique regime where they are constrained to be silent simultaneously. In this talk, I will also introduce time-dependent analysis of HOIs. The MaxEnt models typically assume stationary data. However, it could be that the correlated activity organizes dynamically during behavior and cognition, and this may be independent from spike rates of individual neurons. To resolve this issue, we have been developing the recursive Bayesian estimation method that optimally estimates time-varying neural interactions [Shimazaki et al. ICASSP 2009, PLOS CB 2012, JPCS 2013]. I will introduce a result of the dynamic HOI observed in awake monkey M1, as well as recent development in approximate methods for large-scale state-space analysis of the dynamic neural interactions.
  • Population coding of neurons: Dynamics, higher-order interactions, and mechanisms  [通常講演]
    島崎 秀昭
    Seminar 2016年07月 口頭発表(一般) Queen Mary University Lars Chittka
     
    Population activity of neurons is constrained by their biophysical properties and network architectures. By accurately assessing its statistical structure, and comparing it with model prediction, we gain insights into the underlying networks and their coding principles. The maximum entropy (MaxEnt) principle has been successfully used to describe the population activity from limited amount of data. Conventionally, the MaxEnt model is characterized by parameters for interactions among subset neurons. Recent studies show that neural populations express significant higher-order interactions (HOIs), namely interactions among more-than 2 neurons. However, the previous studies have not identified a key feature that can summarize the seemingly diverse HOIs. Here we examined HOIs of the hippocampal CA3 neurons in cultured slices [Shimazaki et al. Sci Rep 2015], and report that most neurons that expressed HOIs exhibited significantly longer periods of simultaneous silence than predicted by a pairwise MaxEnt model. This simultaneous silence explained ~20% of the entropy related to the HOIs. We also confirmed presence of the structured HOIs predicted from the simultaneous silence. Through two different modeling approaches, we demonstrate that population activity caused by locally shared inputs and nonlinear thresholding reproduces the structured HOIs, and that this structure conveys information of input. This ubiquitous structure of HOIs observed in the activities of both experimental and model neural populations suggests that neurons are operating in a unique regime where they are constrained to be silent simultaneously. In this talk, I will also introduce time-dependent analysis of HOIs. The MaxEnt models typically assume stationary data. However, it could be that the correlated activity organizes dynamically during behavior and cognition, and this may be independent from spike rates of individual neurons. To resolve this issue, we have been developing the recursive Bayesian estimation method that optimally estimates time-varying neural interactions [Shimazaki et al. ICASSP 2009, PLOS CB 2012, JPCS 2013]. I will introduce a result of the dynamic HOI observed in awake monkey M1, as well as recent development in approximate methods for large-scale state-space analysis of the dynamic neural interactions.
  • Exact spike-timing distribution reveals higher-order interactions  [通常講演]
    Shomali SR, Ahmadabad MN, Shimazaki H, Rasuli SN
    CNS*2016 2016年07月 
    It has been suggested that variability in spike patterns of individual neuron is largely due to noisy fluctuations caused by asynchronous synaptic inputs, balanced near the threshold regime [1,2,3]. In this regime, small fluctuations in synaptic inputs to a neuron do cause output spikes; because the membrane potential is maintained below but close enough to the threshold potential. To successfully transfer signals under such noisy conditions, it is proposed that a few relatively stronger synapses and/or an assembly of nearly synchronous ones form “signaling inputs” [4]. Thus one fundamental question is how such relatively strong signaling input modifies the spiking activity of a post-synaptic neuron which receives noisy background input, balanced near the threshold regime. Nonetheless, analytical studies on the effect of the signaling input under such conditions are scarce even with the popular leaky integrate-and-fire (LIF) neuron model. Here we analytically study the impact of a specified signaling input on spike timing of the postsynaptic LIF neuron which receives noisy inputs at the threshold regime. To this end, we first revisit Fokker-Planck analysis of a first spike-timing distribution when the LIF neuron receives noisy synaptic inputs, but no signaling input, at the threshold regime. We then perform perturbation analysis to investigate how a signaling input modifies this first spike-timing distribution. Fortunately, we could solve all terms of perturbation analytically and find the exact first spike-timing distribution of the postsynaptic neuron; it is applicable to not only excitatory but also inhibitory input. This analytical solution allows us to describe the statistics of output spiking activity as a function of background noise, membrane dynamics, and signaling input’s timing and amplitude. The proposed analysis of signaling input provides a powerful framework for studying information transmission, neural correlation, and timing-dependent synaptic plasticity. Among them, we investigate the impact of common signaling inputs on population activities of postsynaptic neurons. Using mixture models based on our analytical first spike-timing distribution, we calculate the higher-order interactions [5] of postsynaptic neurons in different network architectures. Comparing these results with higher-order interactions, measured from experimental data in monkey V1 [6], we try to answer whether one can reveal network architecture, responsible for the ubiquitously observed sparse activities.
  • Simultaneous silence explains structured higher-order interactions of neural population  [通常講演]
    Hideaki Shimazaki
    MONA2 - Modelling Neural Activity 2016年06月 口頭発表(一般) 
    Collective spiking activity of neurons is the basis of information processing in the brain. However, characterizing the population activity is non-trivial because the number of activity patterns combinatorially increases with the number of neurons. To infer the statistical structure of neural activity from limited amount of data, the maximum entropy principle has been successfully applied. Conventionally, the maximum entropy distribution is characterized by interaction parameters of different orders, where the orders refer to the numbers of subset neurons that these parameters constrain. Interactions beyond the 2nd order are collectively termed higher-order interactions (HOIs). Recent studies reported that neural populations express statistically significant HOIs, and they are relevant for information coding. However, the previous studies have not identified a key feature in HOIs that summarizes a principal role of seemingly diverse HOIs. Here we examined HOIs in population activity of the hippocampal CA3 networks in cultured slices [Shimazaki et al. Sci Rep 2015]. To investigate the structure of HOIs, we propose a maximum entropy model that adds a single HOI parameter that accounts for the level of simultaneous silence of all neurons to the previously proposed pairwise maximum entropy model. This single parameter introduces structured HOIs with alternating signs with respect to the order of interactions, namely, negative triple-wise interactions followed by positive quadruple-wise interactions, and then negative quintuple-wise interactions and so on. Using this model, we found that most groups of neurons that expressed HOIs exhibited significantly longer periods of simultaneous silence than predicted by the pairwise maximum entropy model. Indeed, about ~20% of the entropy due to HOIs in these groups was explained by the simultaneous silence. We then directly confirmed presence of the specific alternating structure of HOIs predicted from the SS in the spontaneous activity of the hippocampal neurons. Through a modeling approach, we also demonstrate that population activity caused by correlated inputs and nonlinear thresholding reproduces the same structure of HOIs, and that this structure conveys information of input. The ubiquitous structure of HOI observed in the activities of both experimental and model neural populations suggests that neurons are operating in a unique regime where they are constrained to be silent simultaneously.
  • Toward thermodynamic principles of consciousness  [通常講演]
    Hideaki Shimazaki
    Consciousness club 2016年06月 口頭発表(一般) University of Tokyo Ryota Kanai
     
    Population activity of neurons is constrained by their biophysical properties and network architectures. Thus neurons are statistically dependent each other, and collectively process information. Methods in thermodynamics and statistical mechanics are particularly useful to understand such collective activity of interacting elements. Previous approaches that use models of statistical mechanics assumed stationary neural data. However, it could be that the correlated activity organizes dynamically during behavior and cognition, and this may be independent from spike rates of individual neurons. To resolve this issue, we have been developing a recursive Bayesian estimation method that optimally estimates time-varying neural interactions [1]. I will introduce a result of dynamic (higher-order) interaction of neurons in monkey M1 that is correlated with monkey’s intention to move an arm [1], as well as recent findings on structured higher-order interactions of hippocampal neurons and possible underlying mechanisms [2]. In addition, recent development in approximate methods for large-scale analysis of neural populations will be discussed. The large-scale analysis allows us to characterize dynamic thermodynamic quantities of networks such as entropy, sparsity, and sensitivity. These methods are being developed in part to test ‘Neural Engine Hypothesis’ proposed in [3], my theory toward ‘consciousness’ based on thermodynamic principles. I will demonstrate that neural systems act similarly to a heat engine when an organism actively modulates incoming sensory information as is expected for e.g., attention or reward modulation. This view allows us to quantify the amount of internal computation in a brain, and suggests that an organism is `conscious’ when it acts as an information-theoretic engine.
  • 相関を伴う神経回路網の活動と情報コーディング  [通常講演]
    島崎 秀昭
    ホンダ・リサーチ・インスティチュート 2016年03月 口頭発表(一般)
  • 神経ネットワークの情報コーディング:ダイナミクス・高次相関・メカニズム  [招待講演]
    島崎 秀昭
    第4回東工大若手物性セミナー 2016年02月 公開講演,セミナー,チュートリアル,講習,講義等 
    脳の神経細胞はネットワークを形成して集団として活動することで外界の情報や行動をコードしています.従って神経細胞集団の活動の相関構造を調べることで神経回路網による情報表現を明らかにできると考えられます. 本講演では統計物理学・機械学習領域で用いられる指数型分布族を用いて神経活動の相関を定量化する手法とデータへの適用結果を紹介し,その生成メカニズムを検討します.初めにダイナミックな神経活動を記述する状態空間モデルを導入し,神経細胞の集団的活動を表す高次相関が動的に情報コーディングに関わること示します.次に神経活動に特徴的な高次相関構造がどのようなメカニズムで生成されるのかを数理モデルを用いて検証した結果を紹介します.
  • 動的イジングモデルを用いた神経回路網活動の解析  [招待講演]
    島崎 秀昭
    国際基督教大学 NSフォーラム 2016年02月 公開講演,セミナー,チュートリアル,講習,講義等 
    脳の神経細胞はネットワークを形成し相互に依存しあいながら活動することで外界の情報や行動をコードしています.従って神経細胞間の相関活動を調べることで神経回路網による情報表現を明らかにすることができます. 本講演では統計物理学のイジングモデル(機械学習領域ではボルツマンマシンと呼ばれる)を用いて神経細胞集団の相関活動を定量化する手法とデータへの適用結果を紹介します. これまでの先行研究では定常性を仮定する通常のイジングモデルを用いて培養神経細胞や麻酔下の動物の神経細胞活動の相関構造が明らかにされてきました.しかし覚醒・行動下の動物では刺激や行動に応じて個々の神経細胞の発火頻度と相関構造が動的に変動します.従って通常のイジングモデルをそのまま適用することができません.そこで我々は状態空間モデルに基づく動的イジングモデルを考え,その変動パラメータの逐次ベイズ推定手法を構築しました.講演では本手法をサルの一次運動野神経細胞の同時記録スパイク時系列に適用した結果を紹介し,神経細胞間の高次相関が神経コーディングに関わること示します.また平均場・ベーテ近似等の各種解近似手法を用いて動的イジングモデル解析を大規模化する試みも紹介します.
  • Approximation methods for inferring time-varying interactions of a large neural population  [通常講演]
    Christian Donner, Hideaki Shimazaki
    NIPS 2015 Workshop on `Statistical Methods for Understanding Neural Systems' 2015年12月 ポスター発表 Montreal, Canada
  • 神経回路網の熱力学的考察  [通常講演]
    島崎 秀昭
    2015年11月 口頭発表(一般) 湖邸滋びわこくらぶ 数理モデリング研究会
  • Simultaneous silence explains structured higher-order interactions of neural populations  [招待講演]
    Shimazaki H
    Juelich Research Center 2015年05月 
    Collective spiking activity of neurons is the basis of information processing in the brain. However, characterizing population activity is non-trivial because the number of activity patterns combinatorially increases with the number of neurons. To infer the statistical structure of neural activity from limited amount of data, the maximum entropy principle has been successfully applied. Under this principle, the probability distribution of activity patterns is estimated to be the least structured distribution that is consistent with a set of observed activity statistics. Conventionally the maximum entropy distribution is characterized by interaction parameters of different orders, where the orders refer to the numbers of subset neurons that these parameters constrain. Interactions beyond the 2nd order are collectively termed higher-order interactions (HOIs). Although earlier studies demonstrated that the maximum entropy model that includes up to the 2nd order interactions explains a major variability of population activity, recent studies reported that neural populations express statistically significant HOIs, and they are relevant for information coding. However, the previous studies have not identified a key feature in HOIs that summarizes a principal role of seemingly diverse HOIs. Here we examined HOIs in population activity of the hippocampal CA3 networks in cultured slices [Shimazaki et al. Sci Rep 2015]. To investigate the structure of HOIs, we propose a maximum entropy model that adds a single HOI parameter that accounts for the level of simultaneous silence of all neurons to the previously proposed pairwise maximum entropy model. This single parameter introduces structured HOIs with alternating signs with respect to the order of interactions, namely, negative triple-wise interactions followed by positive quadruple-wise interactions, and then negative quintuple-wise interactions and so on. Using this model, we found that most groups of neurons that expressed HOIs exhibited significantly longer periods of simultaneous silence than predicted by the pairwise maximum entropy model. Indeed, about ~20% of the entropy due to HOIs in these groups was explained by the simultaneous silence. We then directly confirmed presence of the specific alternating structure of HOIs predicted from the SS in the spontaneous activity of the hippocampal neurons. Through a modeling approach, we also demonstrate that population activity caused by correlated inputs and nonlinear thresholding reproduces the same structure of HOIs, and that this structure conveys information of input. The ubiquitous structure of HOI observed in the activities of both experimental and model neural populations suggests that neurons are operating in a unique regime where they are constrained to be silent simultaneously.
  • Simultaneous silence explains structured higher-order interactions of neural populations  [招待講演]
    Shimazaki H
    BCCN Berlin 2015年04月 
    Collective spiking activity of neurons is the basis of information processing in the brain. However, characterizing population activity is non-trivial because the number of activity patterns combinatorially increases with the number of neurons. To infer the statistical structure of neural activity from limited amount of data, the maximum entropy principle has been successfully applied. Under this principle, the probability distribution of activity patterns is estimated to be the least structured distribution that is consistent with a set of observed activity statistics. The maximum entropy distribution is characterized by interaction parameters of different orders, where the orders refer to the numbers of subset neurons that these parameters constrain. Interactions beyond the 2nd order are collectively termed higher-order interactions (HOIs). Although earlier studies demonstrated that the maximum entropy model that includes up to the 2nd order interactions explains a major variability of population activity [Schneidman et al. Nature 2006; Shlens et al. J Neurosci 2006], recent studies reported that neural populations express statistically significant HOIs, and they are relevant for information coding [Ohiorhenuan et. al. Nature 2010; Ganmor et al. PNAS 2011; Shimazaki et al. PLOS CB 2012; Tkacik et al. PLOS CB 2014]. However, the previous studies have not identified a key feature in HOIs that summarizes a principal role of seemingly diverse HOIs. Here we examined HOIs in population activity of the hippocampal CA3 networks in cultured slices [Shimazaki et al. Sci Rep in press]. To investigate the structure of HOIs, we propose a maximum entropy model that adds a single HOI parameter that accounts for the level of simultaneous silence (SS) of all neurons to the previously proposed pairwise maximum entropy model. This single parameter introduces structured HOIs with alternating signs with respect to the order of interactions, namely, positive pairwise interactions followed by negative triple-wise interactions, and then positive quadruple-wise interactions and so on. Using this model, we found that most groups of neurons that expressed HOIs exhibited significantly longer periods of SS than predicted by the pairwise maximum entropy model. Indeed, about ~20% of the entropy due to HOIs in these groups was explained by the single SS term of HOIs. We then directly confirmed presence of the specific oscillatory structure of HOIs predicted from the SS in the population activity of the hippocampal neurons. Through a modeling approach, we also demonstrate that population activity caused by correlated inputs and nonlinear thresholding reproduces the same structure of HOIs, and that this structure conveys information of input. The ubiquitous structure of HOI observed in the activities of both experimental and model neural populations suggests that neurons are operating in a unique regime where they are constrained to be silent simultaneously.
  • Simultaneous silence explains structured higher-order interactions of neural populations  [通常講演]
    Hideaki Shimazaki
    第16回ノンパラメトリック統計解析とベイズ統計 2015年03月 口頭発表(招待・特別)
  • 神経細胞の集団活動の高次統計とメカニズム  [通常講演]
    島崎 秀昭
    第3回ヘテロ・ニューロ・アナリシス研究会 2015年03月 口頭発表(招待・特別)
  • Thomas Sharp, Hideaki Shimazaki, Yoshikazu Isomura, Tomoki Fukai
    Society for Neuroscience 2014 2014年11月 ポスター発表 
    Neurons in layers 2/3 and 5 of rat motor-cortex vary their firing rates and LFP-relative phases during volitional lever-pull tasks. Using a state-space model of spike-time interactions, we analysed the intra- and inter-layer synchrony of these cells in an attempt to understand the behaviour-dependent flow of information in motor cortex. For the whole trial period, at millisecond precision we found strong desynchrony within L5, little synchrony or desynchrony within L2/3, and a peak in inter-layer synchrony at a zero-lag of L5 spike times. Comparing lever-hold and -pull periods we observed that desynchrony was further enhanced in L5 cell pairs that showed concomitant increasing firing rates, and synchrony across layers increased at negative L5 lags, that is, when we subtracted 2, 4, 6 or 8 milliseconds from L5 spike times. These results suggest that the two layers share significant common inputs and that information flows from the shallow to deep layers specifically during arm movement.
  • Theoretical study on spike-timing probability in a pair of pre-post synaptic neurons  [通常講演]
    Safura Rashid Shomali, Majid Nili Ahmadabadi, Hideaki Shimazaki, S Nader Rasuli
    Neuro2014 2014年09月 ポスター発表 
    Neurons in the cortex are embedded in intricate networks and produce spikes in response to bombardment of stochastic, balanced synaptic inputs from excitatory and inhibitory presynaptic neurons when processing information. While the noisy synaptic inputs may obscure the effect of individual presynaptic neurons on postsynaptic spike generation, the output spikes are certainly not random, but can be controlled by features of selected input spike patterns buried in the noisy inputs. Although fundamentals of neuronal information transmission are described by such interactions between the input statistics and spiking nonlinearity, analytical studies on the effect of presynaptic inputs on generation of postsynaptic spikes under the balanced conditions are scarce; even in a simple leaky integrate-and-fire (LIF) neuron model. In this study, we theoretically investigate the effect of a single presynaptic neuron on spike timing of a post-synaptic neuron using the LIF model receiving noisy balanced inputs. Through diffusion approximation of the model and some linear analysis on the corresponding Fokker-Planck equation, we provide the probability distribution of postsynaptic spike-timing conditioned on a pre-synaptic neuron’s spike timing. In particular, we analytically demonstrate the change in the post-synaptic neuron’s spiking probability caused by individual or synchronous presynaptic inputs. The result of this study is expected to be useful to analyze capacity of a LIF model neuron receiving balanced inputs, and further investigate synaptic weights’ dynamics; that are continuously strengthened or weakened by causal plasticity rules.
  • Behaviour- and layer-dependent synchrony in motor cortex during volitional arm movement  [通常講演]
    Thomas Sharp, Hideaki Shimazaki, Yoshikazu Isomura, Tomoki Fukai
    Neuro2014 2014年09月 ポスター発表 
    Neurons in layers 2/3 and 5 of rat motor-cortex vary their firing rates and LFP-relative phases during volitional lever-pull tasks. Using a state-space model of spike-time interactions, we analysed the intra- and inter-layer synchrony of these cells in an attempt to understand the behaviour-dependent flow of information in motor cortex. For the whole trial period, at millisecond precision we found strong desynchrony within L5, little synchrony or desynchrony within L2/3, and a peak in inter-layer synchrony at a zero-lag of L5 spike times. Comparing lever-hold and -pull periods we observed that desynchrony was further enhanced in L5 cell pairs that showed concomitant increasing firing rates, and synchrony across layers increased at negative L5 lags, that is, when we subtracted 2, 4, 6 or 8 milliseconds from L5 spike times. These results suggest that the two layers share significant common inputs and that information flows from the shallow to deep layers specifically during arm movement.
  • Sharp T, Shimazaki H, Isomura Y, Fukai T
    Workshop on data mining in neuroscience 2014年05月 口頭発表(一般) National Institute of Informatics, Tokyo
  • Bimodal distributions of local phase variables in natural images  [通常講演]
    HaDi MaBouDi, Hideaki Shimazaki, Hamid Soltanian-Zadeh, Shun-ichi Amari
    The 2014 VSS Annual Meeting 2014年05月 ポスター発表 St. Pete Beach, Florida 
    Natural scenes contain rich information in their local phase structure compared to their amplitudes. Nevertheless, conventional models of visual systems represent the natural images using superposition of receptive fields (RFs) based on the information contained only in the amplitudes. As a result, the redundancy in the images is not completely removed. Typically, the responses of the RFs after training exhibit circular dependencies. Thus recent studies suggested decomposing images using amplitude and phase coefficients of a complex representation. However, these studies assume uniform phase distributions. Here, we report the presence of structured bimodal distributions for the phase variables of the complex RFs responding to natural scenes, suggesting that the uniform distribution is insufficient to reduce the redundancy of natural scenes. To show this, we first construct a complex RF defined as a pair of Gabor-like RFs possessing the same features such as scales, orientations, and frequencies, but are in quadrature phase. Next, we obtain a phase distribution of its responses to patches selected from whitened natural scenes. The phase distribution was then fitted by a mixture of two von Mises distributions and a uniform circular distribution, using the expectation-maximization algorithm developed for this study. Finally, we apply the complex RFs possessing different features to the natural image patches to investigate variation of the phase distributions with these features. The analysis revealed a half of the complex RFs exhibited bimodal distributions. The distances between two peaks of the distributions were about 180 degree. Importantly, the shape of the distributions significantly varied when the scale or frequency of the complex RFs changed: The complex RFs possessing low frequencies or small scales yielded the bimodal distributions. Our results suggest that the redundancy in the natural images can be further removed if we consider bimodal phase distributions, in particular, for low-frequency / small-scale complex RFs.
  • State-space analysis of higher-order interactions in parallel event sequences  [通常講演]
    Shimazaki H, Amari S, Brown EN, Grün S
    第15回ノンパラメトリック統計解析とベイズ統計 2014年03月 口頭発表(招待・特別) 慶応義塾大学三田キャンパス
  • 高次相関を伴う神経回路網活動:局所回路の計算原理を求めて  [通常講演]
    島崎 秀昭
    神経科学と脳科学の対話4 2014年03月 口頭発表(一般) 立川,東京 統計数理研究所
  • Statistical inference for directed phase coupling in neural oscillators  [通常講演]
    HaDi MaBouDi, Hideaki Shimazaki, Mehdi Abouzari, Shun-ichi Amari, Hamid Soltanian-Zadeh
    Computational and Systems Neuroscience (Cosyne) 2014 2014年02月 ポスター発表 Salt Lake City, USA 
    Phase coupling and synchronization have been an important concept in the study of oscillations of neural systems for the past few decades. The phase coupling of various neuronal measurements was observed within local activity of cortical and subcortical areas and even between activities in distant brain regions [1]. Nevertheless, previous methods for estimating phase couplings are limited. Both standard phase correlation and phase coherence compute static pairwise dependency of two oscillators, therefore are limited to uncover underlying couplings of multiple oscillators, and are blind to their underlying dynamics. Recently, Cadieu and Koepsell [2] provided a model-based method that allows us to directly and simultaneously estimate underlying undirected couplings of multiple oscillators using the Kuramoto model of an oscillator network. However, this and other previous methods cannot reveal causal relations between the oscillators. Here we propose a method to estimate the causality in the phase space from measurements of neural oscillations. A new probabilistic method was developed to estimate directed coupling parameters of the phase oscillators from noisy multivariate circular time-series data. In this method, we constructed a probabilistic description of the Kuramoto model, and developed an algorithm to estimate the coupling parameters under the maximum likelihood principle. We demonstrate that the proposed method recovers the underlying direction and weights of couplings between noisy dynamic oscillators while previous approaches by the phase correlations and the Granger causality were unable to recover these values correctly. We stress that the method performs well on a relatively large number of oscillators even in the presence of noise. The estimated directed graphs provide us useful tools to uncover causal relations of cortical networks, such as casual interactions of local micro-circuitries, and extract their network topology. [1] Buzsaki, G. and et al. (2013) Neuron. [2] Cadieu and Koepsell (2010) Neural Comput.
  • Statistical inference for directed phase coupling in neural oscillators. Computational and Systems Neuroscience  [通常講演]
    HaDi MaBouDi, Hideaki Shimazaki, Mehdi Abouzari, Shun-ichi Amari, Hamid Soltanian-Zadeh
    Cosyne 2014, Salt Lake City, USA. 2014年02月 ポスター発表
  • Structured higher-order interactions explain the simultaneous silence of neural populations  [通常講演]
    Shimazaki H, Sadeghi K, Ikegaya Y, Toyoizumi T
    脳と心のメカニズム 第14回 冬のワークショップ 2014年01月 ポスター発表
  • The simultaneous silence of neurons explains structured higher-order interactions in ensemble spiking activity  [通常講演]
    Shimazaki H, Sadeghi K, Ikegaya Y, Toyoizumi T
    Society for Neuroscience (SfN) 2013 2013年11月 ポスター発表 San Diego, USA 
    Collective spiking activity of neurons is the basis of information processing in the brain. Sparse neuronal activity in a population of neurons limits possible spiking patterns and, thereby, influences the information content conveyed by each pattern. However, characterization of ensemble spiking activity has been mostly limited to lower-order interactions, such as pairwise interactions, because the number of parameters required to describe higher-order interactions (HOIs) combinatorially increases with the number of neurons. To investigate the structure of HOIs, we propose a new heuristic model that adds a single HOI parameter that accounts for the level of simultaneous silence of multiple neurons to the previously proposed pairwise interaction model. This single parameter introduces structured HOIs with alternating signs with respect to the order of interactions, namely, positive pairwise interactions followed by negative triple-wise interactions, and then positive quadruple-wise interactions and so on. Using this model, we tested if simultaneously recorded groups of neighboring neurons in the hippocampal CA3 area express excess simultaneous silence under spontaneous activity. Most groups of neurons whose entropy in ensemble activity was not fully characterized by firing rates and pairwise correlations exhibited significantly longer periods of simultaneous silence than predicted by the pairwise interaction model. Indeed, about 1/3 of the entropy due to HOIs in these groups was explained solely by the simultaneous silence of neurons. Through a modeling approach, we show that simultaneous silence is a generic outcome of neurons that have spiking nonlinearity and correlated input. We demonstrate that ensemble activity of binary neurons that receive correlated Gaussian inputs exhibit the structured HOIs predicted by the simultaneous silence. We discuss the implication of the simultaneous silence for neural information coding.
  • State-space analysis of time-varying higher-order interactions: its applications to neuroscience  [招待講演]
    Shimazaki H
    ELC International Meeting on ''Inference, Computation, and Spin Glasses'' (ICSG2013) 2013年07月 口頭発表(招待・特別) Sapporo, Japan
  • Shimazaki H
    Workshop on statistical analysis of neurophysiological and clinical data 2013年07月 口頭発表(招待・特別) Kyoto, Japan Kyoto University
  • Shimazaki H, Amari S, Brown EN, Grün S
    Modeling Neural Activity: Statistics, Dynamical Systems, and Networks 2013年06月 口頭発表(招待・特別) Lihue, Hawaii, USA 
    Neurons embedded in a network exhibit correlated spiking activity, and can produce synchronous spikes with millisecond precision. It was reported that the correlated activity organizes dynamically during behavior and cognition, independently from spike rates of individual neurons. However, in order to detect the dependency of multiple neurons, it may be necessary to estimate their `higher-order' interactions that can not be inferred from simultaneous activities of only pairs of neurons from the group. To estimate the time-varying higher-order neuronal interactions, we recently developed a method that combines the model of higher-order interactions with a state-space method. We applied this method to three neurons recorded simultaneously from the primary motor cortex of a monkey engaged in a delayed motor task (data from Riehle et al., Science, 1997). We found that depending on the behavioral demands due to the task these neurons dynamically organized into a group characterized by the presence of higher-order (triple-wise) inter??action. There was, however, no noticeable change in he firing rates of the involved neurons. The analysis demonstrates that time-varying higher-order analysis allows us to detect subsets of correlated neurons that may belong to a larger set of neurons comprising a neuronal cell assembly. Ref: Shimazaki et al., PLoS Comp. Biol. (2012) 8(3): e1002385
  • The simultaneous silence of neurons explains higher-order interactions in ensemble spiking activity  [通常講演]
    Shimazaki H, Sadeghi K, Ikegaya Y, Toyoizumi T
    Neuro2013 2013年06月 ポスター発表 Kyoto, Japan 
    Collective spiking activity of neurons is the basis of information processing in the brain. Sparse neuronal activity in a population of neurons limits possible spiking patterns and, thereby, influences the information content conveyed by each pattern. However, because of the combinatorial explosion of the number of parameters required to describe higher-order interactions (HOIs), the characterization of neuronal interactions has been mostly limited to lower-order interactions, such as pairwise interactions. Here, we propose a new model that characterizes population-spiking activity by adding a single parameter to the previously proposed pairwise interaction model. This parameter describes the fraction of time a group of neurons is simultaneously silent, which can be alternatively expressed as a specific combination of HOIs. We apply our model to groups of neighboring neurons that are simultaneously recorded from spontaneously active slice cultures from the hippocampal CA3 area. Most groups of neurons that are not adequately explained by the pairwise interaction model exhibit significantly longer periods of simultaneous silence than the chance level expected from firing rates and pairwise correlations, demonstrating that simultaneous silence is a common property coded by HOIs. To confirm that the simultaneous silence is also a major property, we systematically obtained a one-dimensional data-driven HOI term that is asymptotically optimal when added to a pairwise-interaction model. This analysis exhibited the structured HOIs expected from the simultaneous silence of neurons, i.e., positive pairwise interactions are followed by negative triple-wise interactions, and then positive quadruple-wise interactions. These results suggest that seemingly complex HOIs can be explained by simultaneous silence of multiple neurons. We discuss the implication of simultaneous silence for our understanding of the underlying circuit architecture and information coding.
  • Shimazaki H
    The 3rd Mathematical Neuroscience Workshop in School of Mathematics 2013年03月 口頭発表(招待・特別) Tehran, Iran Institute for Research in Fundamental Sciences (IPM)
  • Shimazaki H
    The 3rd Mathematical Neuroscience Workshop in School of Mathematics 2013年03月 口頭発表(招待・特別) Tehran, Iran Institute for Research in Fundamental Sciences (IPM)
  • 島崎 秀昭
    玉川大学脳科学若手の会 第88回談話会 2013年02月 口頭発表(招待・特別) 玉川大学,東京 玉川大学脳科学若手の会
  • Shimazaki H, Sadeghi K, Ikegaya Y, Toyoizumi T
    神経科学と統計科学の対話3 2013年02月 統計数理研究所 ,立川,東京 統計数理研究所
  • Shimazaki H, Sadeghi K, Ikegaya Y, Toyoizumi T
    Workshop on statistical aspects of neural coding 2012年11月 口頭発表(招待・特別) Kyoto University & Ritsumeikan University Kyoto University & Ritsumeikan University
  • Shimazaki H
    Workshop on neural information flow 2012年06月 口頭発表(招待・特別) Kyoto, Japan Kyoto University
     
    Neurons embedded in a network are correlated, and can produce synchronous spiking activities with millisecond precision. It is likely that these correlated activities organize dynamically during behavior and cognition, and this may be independent from spike rates of individual neurons. Consequently current analysis tools must be extended so that they can directly estimate time-varying neural interactions. The log-linear model is known to be useful for analysis of the correlated spiking activities but is limited to stationary data. In our approach, we developed a `state-space log-linear model’ that can estimate ever-changing neural interactions: this method is an extension of the familiar Kalman filter which can track system’s parameters as used in, e.g., automotive navigation systems. We applied this method to three neurons recorded from the primary motor cortex of a monkey engaged in a delayed motor task (data from Riehle et al., Science,1997). We found that depending on the behavioral demands of the task these neurons dynamically organized into a group which was characterized by the presence of higher-order (triple-wise) interaction. There was, however, no noticeable change in their firing rates. These results demonstrate that time-varying higher-order analysis allows us to detect subsets of correlated neurons that may belong to a larger set of neurons comprising a cell assembly. This is a collaboration work with Shun-ichi Amari (RIKEN), Emery N. Brown (MIT), and Sonja Gruen (Julich). Original paper: Shimazaki et al., PLoS CB 8(3): e1002385
  • ヒストグラム・カーネル密度推定の神経スパイクデータへの適用:理論と実践  [通常講演]
    島崎 秀昭, 篠本 滋
    第13回ノンパラメトリック統計解析とベイズ統計 2012年03月 口頭発表(招待・特別) 慶應義塾大学三田キャンパス
  • The simultaneous silence of neurons explains higher-order interactions in ensemble spiking activity  [通常講演]
    Shimazaki H, Sadeghi K, Ikegaya Y, Toyoizumi T
    Computational and Systems Neuroscience (Cosyne) 2012 2012年02月 ポスター発表 Salt Lake City, USA
  • 神経細胞の高次スパイク相関:状態空間モデルによる解析  [通常講演]
    島崎 秀昭
    京都大学理学部物理学第一教室 非線形セミナー 2012年02月 公開講演,セミナー,チュートリアル,講習,講義等 京都 京都大学
     
    神経細胞集団の同期的なスパイク発火活動が神経系の情報処理に重要な役割を果たしているとする仮説が提案されています.実際,個々の神経細胞の発火頻度だけでなく,2つの神経細胞の同時刻スパイク相関が刺激・行動・脳の内部状態に関連して生じることが報告されています.一方で近年,2つ以上の細胞集団の同 期発火活動の特徴を明らかにしたいという要求から,2次相関解析では見過ごされてしまう神経細胞間の高次の相互作用に注目が集まっています.しかし,刺激・行動に応じて生成される可能性のある動的な高次スパイク相関構造を直接推定する手法はありませんでした.今回,私達は対数線形モデルを観測モデルとするマルチニューロンスパイクデータの状態空間モデルを構築し,遅延課題遂行中のサルの一次運動野神経細胞の同時記録スパイク時系(Riehle et al., Science,1997)に適用しました.その結果,運動開始の合図までの準備期間(腕が動いていない状態)の神経活動において,運動開始に対するサルの内部期待と関連して,発火頻度とは独立に3次スパイク相関が有意に上昇する活動が見られました.本研究により,行動下の神経スパイクデータから細胞集団の動的な協調活動の存在を検出し,動物の認知・行動に果たす役割を明らかにするための解析技術が実用段階に達したと考えています. 本研究は甘利俊一(理研), Emery N. Brown(MIT), Sonja Gruen(Julich)(敬称略)との共同研究です. 参考文献 Shimazaki H., Amari S., Brown E. N., and Gruen S., State-space analysis of time-varying higher-order spike correlation for multiple neural spike train data. PLoS Computational Biology in Press
  • 神経スパイク解析における状態空間モデル,GLMの応用  [通常講演]
    島崎 秀昭, 小山 慎介
    統計科学と神経科学の対話2 2011年12月 口頭発表(招待・特別) 立川,東京 統計数理研究所
  • 脳の高次機能と動的高次スパイク相関:状態空間モデルによる解析  [通常講演]
    島崎 秀昭
    統計神経科学ミニワークショップ 2011年09月 立川,東京 統計数理研究所
     
    神経細胞集団の同期的スパイク発火活動が神経系の情報処理に重要な役割を果たしているとする仮説が古くから提唱されています.この仮説に従えば,個々の神経細胞の発火頻度だけでなく,スパイク時系列間の同時刻相関構造にも外界情報が符号化されている可能性があります.特に,外部刺激等を起因とする共通入力を受けた細胞集団の発火活動は2次相関だけでは表現できない可能性があり,高次スパイク相関と外部刺激・動物の行動との関係を明らかにすることが求められています.スパイク高次相関は対数線形モデルを用いて情報幾何の枠組みで定義することができます.行動に応じて生成される動的な高次スパイク相関構造を推定するために,私達は対数線形モデルを観測モデルとする状態空間モデルを考え,遅延課題中のサルの運動野神経細胞の同時記録スパイク時系列に適用しました. この結果,運動開始の合図があるまでの準備期間(腕が動いていない状態)の神経活動において,行動開始に対するサルの内部期待の高まり(運動開始信号の不確かさの減少)と共に,発火頻度とは独立に3次スパイク相関が上昇する活動が見られました.スパイク高次相関と脳の高次機能との関係を示唆する始めての報告です.高次スパイク相関の検出には周辺尤度比(ベイズ因子)を用い,有意水準を決定するためにサロゲートデータ(対象の高次相関のみが存在しないという帰無仮説の下で生成されたデータ)を作成・使用しました. 一連の研究により,神経スパイクデータから細胞集団の動的な協調活動を検出し, 動物の認知・行動に果たす役割を明らかにするための解析技術が実用段階に達したと考えています. 本研究は甘利俊一(理研), Emery N. Brown(MIT), Sonja Gruen(Julich)(敬称略)との共同研究です.
  • Constructing a joint time-series model of continuous and Bernoulli/Poisson processes using a copula  [通常講演]
    Shimazaki H, Brown EN
    Computational and Systems Neuroscience (Cosyne) 2011 2011年02月 ポスター発表 Salt Lake City, USA 
    Neurophysiologists frequently record neural spiking activity along with local field potentials (LFP), electroencephalography (EEG) or electrocorticographic (ECoG) measurements. Although the simultaneous recordings characterize the properties of the system under investigation, the analyses of the point processes (spiking activity) and the continuous signals (LFP, EEG and ECoG) are conducted separately. If the two types of signals are providing information about the same process then it seems reasonable to develop methods to analyze them simultaneously. Nevertheless, to our knowledge, there have been no attempt to build a joint distribution of the hybrid time-series signals. This is probably because an appropriate mathematical theory to combine continuous and discrete signals has not spread to the Neuroscience community. In this contribution, to model the hybrid signals, we construct a joint model of continuous and Bernoulli/Poisson processes by using a `copula' function: a function that allows us to combine continuous and discrete random variables, and make their joint distribution. Using a copula function, the dependence of the joint processes (both across-time and across-signals) can be modeled independently from their marginal distributions. Following this framework, we first construct multivariate continuous-valued Markov processes that marginally obey non-Gaussian continuous distributions. We then generate a Bernoulli process (for discrete-time) or a Poisson point process (for continuous-time) given the continuous signals by using a novel Bernoulli rate or a conditional intensity function written by a copula function. For empirically observed signals, their dependency can be quantified by estimating the copula using either a parametric or non-parametric method. The joint time-series model by a copula thus provides a quantitative method to assess the dependence of a spike signal and various types of continuous signals recorded simultaneously.
  • Shimazaki H
    Neurostatistics Working Group Seminar 2010年12月 口頭発表(招待・特別) Boston, USA Dept. of Biostatistics, Harvard University
     
    Neurons in the brain communicate with each other using pulsed electrical discharges, known as spikes. Spike data recorded from an awake animal is subject to dynamic change in relation to a stimulus presented to the animal, animal's behavior or an internal state of the brain. In this talk, I will present statistical tools for analyzing the non-stationary (A) single and (B) multiple neural spike data. (A) In electrophysiological studies, neurophysiologists often investigate the discharge rate of single neurons. The peri-stimulus time histogram (PSTH), and the spike density function (SDF), a convolution of a kernel function with spike trains, are classical tools used to examine dynamics of the single neuron spike-rate. However, the histogram bin-width or the kernel bandwidth, which critically determines the shape of the PSTH/SDF, has to be subjectively selected. In this talk, I will present objective methods for optimizing the bin-width [1] and bandwidth [2]. We demonstrate that the SDF exhibits superior fitting performance to the PSTH. Furthermore, the performance of the classical SDF becomes comparable to, or even better than, other state-of-the-art rate estimation tools, provided that the bandwidth is optimized as presented in the talk. (B) Due to the technological advances in multiple-neuron recordings, statistical methods for estimating multiple neural spike correlations are increasingly becoming important. The spike correlation occurring dynamically in relation to behavioral context is discussed as an indicator of active cell assemblies [Hebb, 1949], and thus relevant for information processing. To trace dynamic assembly activities, we recently developed a method that - in contrast to existing analysis methods - allows to estimate simultaneously dynamic spike-rates and higher-order spike correlations by means of a state-space analysis [3]. Based on this model, we formulate a Bayesian hypothesis test on the latent time-varying correlations using the Bayes factor in combination with a surrogate method. The proposed analysis methods allow us to detect assemblies momentarily activated in association e.g. with behavioral events. [1] Shimazaki and Shinomoto. A method for selecting the bin size of a time histogram. Neural Comput (2007) 19(6), 1503-1527 [2] Shimazaki and Shinomoto. Kernel Bandwidth Optimization in Spike Rate Estimation. J Comput Neurosci (2010) 29 (1-2), 171-182 [3] Shimazaki, Amari, Brown, and Gruen. State-space Analysis on Time-varying Correlations in Parallel Spike Sequences. Proc. IEEE ICASSP2009, 3501-3504. * Statistical tools are available at http://2000.jukuin.keio.ac.jp/shimazaki and http://www.ton.scphys.kyoto-u.ac.jp/~shino/toolbox/english.htm
  • Shimazaki H
    Workshop on spatio-temporal neuronal computation 2010年09月 口頭発表(招待・特別) Kyoto, Japan Kyoto University
  • Characterizing neuronal firing with the rate and the irregularity  [通常講演]
    Shinomoto S, Shimazaki H, Shimokawa T
    Neuro 2010 2010年09月 口頭発表(一般) Kobe, Japan 
    Neurons, or nerve cells in the brain, communicate with each other using stereotyped electric pulses, called spikes. Here we wish to introduce methods we recently developed for characterizing the spike train. Firstly, it is believed that neurons convey information mainly through the frequency of the transmitted spikes, called the firing rate. Any rate estimation tool such as the peri-stimulus time histogram (PSTH) and the kernel density estimation has a parameter, such as the binsize of a time histogram and the bandwidth of a kernel smoother that controls the jaggedness of the estimate. The estimated rate may become highly fluctuating if the binsize or bandwidth is small, and constant in the opposite limit. We introduce here the methods for selecting the most suitable parameter from the spike count statistics alone, so that the resulting bar or line graph time histogram best represents the unknown underlying spike rate [1,2].Secondly, neurons may communicate some information through the finer temporal patterns of the spikes. We found that neurons exhibit stable firing patterns that can be characterized as regular, random, and bursty, and these characteristics are strongly correlated to the functions of cortical areas [3]. We wish to introduce methods we have developed for characterizing a sequence of spikes in terms of the irregularity in addition to the rate of firing [4,5].[1] H. Shimazaki and S. Shinomoto, Neural Comput (2007) 19:1503-1700. [2] H. Shimazaki and S. Shinomoto, J Compt Neurosci (2010) in press.[3] S. Shinomoto et al. PLoS Comput Biol (2009) 5:e1000433. [4] T. Shimokawa and S. Shinomoto, Neural Comput (2009) 21:1931-1951. [5] T. Shimokawa, S. Koyama, and S. Shinomoto, J Comput Neurosci (2010) in press
  • Analysis of subsets of higher-order correlated neurons based on marginal correlation coordinates  [通常講演]
    Shimazaki H, Amari S, Brown EN, Grün S
    Computational and Systems Neuroscience (Cosyne) 2010 2010年02月 Salt Lake City, USA
  • 島崎秀昭, 甘利俊一, Emery Brown, Sonja Gruen
    日本神経回路学会 第19回全国大会 2009年09月 口頭発表(一般) 仙台 日本神経回路学会
     
    To trace active assembly activities of neurons from multiple neural spike data, we developed a method for estimating dynamic spike correlations by means of a state-space analysis. Discretized parallel spike sequences are modeled as a non-stationary multivariate Bernoulli process using a log-linear function (a multinomial logit), which introduces states of higherorder interaction terms. We constructed a recursive Bayesian filter/smoother to estimate time-varying loglinear parameters, and an optimization method of hyper-parameters under the likelihood principle.
  • Histogram binwidth and kernel bandwidth selection for the Spike-rate estimation  [通常講演]
    Shimazaki H, Shinomoto S
    CNS2009 2009年07月 ポスター発表 Berlin, Germany 
    Histogram and kernel methods have been used as standard tools for capturing the instantaneous rate of neuronal spike discharges in the neurophysiology community. These methods are left with one free parameter that determines the smoothness of the estimated rate, namely a binwidth or bandwidth. In most of the neurophysiology literature, however, the binwidth or bandwidth that critically determines the goodness of the fit of the estimated rate to the underlying rate has been selected by individual researchers in an unsystematic manner. Recently, we established a method for selecting the histogram binwidth [1] as well as the kernel bandwidth [2], with which the estimated rate best approximates the unknown underlying rate. The resolution of the optimized estimated rate increases, or the optimal bin/band-width decreases, with the number of spike sequences sampled. It is notable that the optimal bin/band-width diverges if only a small number of experimental trials are available from a moderately fluctuating rate process [3]. In this case, any attempt for characterizing the underlying spike rate will lead to spurious results. To assist those who are confronted with such paucity of data, we developed a method that can suggest how many more trials are needed until the set of data can be analyzed with the required resolution.
  • Bayes factor analysis for detection of time-dependent higher-order spike correlations  [通常講演]
    Shimazaki H, Amari S, Brown EN, Grün S
    CNS2009 2009年07月 ポスター発表 Berlin, Germany 
    Precise spike coordination in the spiking activities of a neuronal population is discussed as an indication of coordinated network activity in form of a cell assembly relevant for information processing. Supportive evidence was provided by the existence of excess spike synchrony occurring dynamically in relation to behavioral context [1-3]. These findings are based on the measured dependence of multiple neurons against the null-hypothesis of full independence. However, neurons jointly involved in assemblies may express higher-order correlations (HOCs) in their spiking activities [4]. By characterizing the spatio-temporal pattern of parallel spikes with the HOCs, one may elucidate assembly activities and possibly their behavioral relevance. To describe the HOCs in parallel spike trains, the log-linear model is an useful model because it provides a well-defined measure of correlation based on information geometry [5]. Former studies on HOCs performed a regression analysis on parallel spike trains using either a full log-linear model [6] or a log-linear model containing up to pairwise interaction only (maximum entropy model) [7]. The existing approaches, however, assume stationarity, a condition that is typically not fulfilled in neuronal spike data from awake behaving animals. Recently, we established a method for estimating the dynamics of HOCs by means of a state-space analysis [8] with a log-linear observation model to trace active assemblies [Abstracts in SAND4, NIPS08 Workshop, and Cosyne09, [9]]. However, presentation of the smoothed posterior estimates only may mislead neurophysiologists to presume the existence of the HOC at the moment when no or weak evidence is available. Furthermore, the method did not provide a statistic to detect an assembly in which cells are jointly connected through multiple correlations. In this contribution, we investigate the method of the Bayesian hypothesis testing to answer which compositions of parallel spikes exhibit the joint HOCs, and if they do, when those HOCs appear. We computed the Bayes factor[10] of temporally local spike observation to gain evidence of positive joint HOCs of a specific set of parallel spike sequences against negative HOCs by using filter and one-step prediction odds. The proposed method may be useful to detect the dynamic assembly activities, their composition and behavioral relevance when applied to simultaneous recordings of neuronal activity of behaving animal.
  • State-space model of dynamic correlations in parallel spike sequences  [招待講演]
    島崎 秀昭
    京都大学理学研究科物理学第一教室 非線形動力学セミナー 2009年06月 公開講演,セミナー,チュートリアル,講習,講義等 
    To trace dynamic assembly activities of neurons from multiple neural spike data, we recently established a method for simultaneously estimating time-varying spike rate and correlations by means of a state-space analysis [Shimazaki, Amari, Brown, and Gruen: State-space Analysis on Time-varying Correlations in Parallel Spike Sequences. Proc. IEEE ICASSP, 3501-3504, 2009]. Precise spike coordination in spiking activities of multiple neurons is discussed as an indication of coordinated network activity in form of a cell assembly [Hebb, 1949] realizing information processing. Such collective activities may be dynamically organized depending on the behavioral context. Supportive evidence was provided by the existence of excess spike synchrony occurring dynamically in relation to behavioral demands [e.g. Riehle et al., Science, 1997]. These findings are based on the measured dependence of multiple neurons against the null-hypothesis of full independence. However, spiking activities of neurons jointly processing information may be properly expressed if higher-order correlations (HOCs) are considered. The log-linear model [Amari, IEEE Trans Inf Theory, 2001] provides a well-defined measure of HOC based on information geometry. Previous work based on the log-linear model [e.g. Martignon, Biol Cybern, 1995], however, assumes stationarity, a condition that is typically not fulfilled in neuronal spike data in particular from awake behaving animals. As a solution, we constructed a state-space model of dynamic correlations using the log-linear observation model, and developed a recursive Bayesian filter/smoother for estimating the time-varying log-linear parameters. Examining the relation of time-varying HOCs to behavioral aspects may provide us with new insights into the dynamics of assembly activities, their composition and behavioral relevance. This is a collaboration work of Hideaki Shimazaki1, Shun-ichi Amari1, Emery N. Brown2,3, Sonja Gruen1 1 RIKEN Brain Science Institute, Wako-shi, Saitama, Japan 2 Anesthesia and Critical Care, Massachusetts General Hospital, Boston, MA, USA 3 Harvard-MIT Division of Health Sciences and Technology, Cambridge, MA, USA
  • Estimating time-varying spike correlations from parallel spike sequences  [通常講演]
    Shimazaki H, Amari S, Brown EN, Grün S
    German-Japanese Workshop "Computational and Systems Neuroscience" 2009年05月 ポスター発表 Berlin, Germany 
    Precise spike coordination in the spiking activities of a neuronal population is discussed as an indication of coordinated network activity in form of a cell assembly relevant for information processing. Supportive evidence was provided by the existence of excess spike synchrony occurring dynamically in relation to behavioral context [e.g. Riehle et al. Science 278, 1950-1953, 1997]. These findings are based on the measured dependence of multiple neurons against the null-hypothesis of full independence. However, spiking activities of neurons jointly involved in assemblies may express complex spike correlations including higher-order statistical dependency [Grün et al. Lect Notes Comput Sci 5286, 96-114, 2008]. By characterizing the higher-order correlations (HOCs) in the observed spatio-temporal pattern of parallel spike, one may elucidate compositions of assemblies, and their behavioral relevance. The log-linear model is an useful model to describe the correlations because it provides a well-defined measure of HOCs based on information geometry [Amari IEEE Trans Inf Theory 47, 1701-1711, 2001; Nakahara and Amari, Neural Comput 14, 2269-2316, 2002]. Former studies on HOCs were based on a regression analysis using either a full log-linear model [Martignon et al. Biol Cybern 73, 69-81, 1995], or a log-linear model containing up to pairwise interaction only (maximum entropy model) [Schneidman et al. Nature 440, 1007-1012, 2006]. The existing approaches, however, assumed stationarity, a condition that is typically not fulfilled in neuronal spike data from awake behaving animals. To trace active assemblies in non-stationary spike data, we recently established a method for simultaneously estimating time-dependent spike rate and HOCs by means of a statespace analysis combined with a log-linear observation model [Shimazaki H, Amari S, Brown EN, and Grün S: State-space Analysis on Time-varying Correlations in Parallel Spike Sequences. Proc. IEEE ICASSP, 2009]. In this method, discretized parallel spike trains are modeled by a conditionally independent multivariate Bernoulli process using a log-linear (or a multivariate logit) link function. We developed a nonlinear recursive filtering/smoothing algorithm for estimating the time-varying log-linear parameters by applying a log-quadratic approximation to its posterior distribution. The time-scales of the smoothed estimate of each parameter and their covariation are automatically optimized via the EM-algorithm under the maximum likelihood principle. The proposed multivariate state-space model of parallel spike trains will be applicable to the multiple neural spike data obtained from an awake behaving animal. Examining the relation of time-varying HOCs to the animal's behavior may provide us with new insights into the dynamics of assembly activities, their composition, and behavioral relevance.
  • State-space Analysis on Time-varying Correlations in Parallel Spike Sequences  [招待講演]
    Shimazaki H, Amari S, Brown EN, Grün S
    ICASSP2009 Special Session on `Signal Processing for Neural Spike Trains' 2009年04月 口頭発表(招待・特別) Taipei, Taiwan IEEE
     
    A state-space method for simultaneously estimating timedependent rate and higher-order correlation underlying parallel spike sequences is proposed. Discretized parallel spike sequences are modeled by a conditionally independent multivariate Bernoulli process using a log-linear link function, which contains a state of higher-order interaction factors. A nonlinear recursive filtering formula is derived from a logquadratic approximation to the posterior distribution of the state. Together with a fixed-interval smoothing algorithm, time-dependent log-linear parameters are estimated. The smoothed estimates are optimized via EM-algorithm such that their prior covariance matrix maximizes the expected complete data log-likelihood. In addition, we perform model selection on the hierarchical log-linear models to avoid overfitting. Application of the method to simultaneously recorded neuronal spike sequences is expected to contribute to uncover dynamic cooperative activities of neurons in relation to behavior.
  • Detection of non-stationary higher-order spike correlation  [通常講演]
    Shimazaki H, Amari S, Brown EN, Grün S
    Computational and Systems Neuroscience (Cosyne) 2009 2009年02月 ポスター発表 Salt Lake City, USA 
    Precise spike coordination in the spiking activities of a neuronal population is discussed as an indication of coordinated network activity in form of a cell assembly relevant for information processing. Supportive evidence for its relevance in behavior was provided by the existence of excess spike synchrony occurring dynamically in relation to behavioral context [e.g. Riehle et. al., Science (278) 1950-1953, 1997]. This finding was based on the null-hypothesis of full independence. However, one can assume that neurons jointly involved in assemblies express higher-order correlation (HOC) between their activities. Previous work on HOC assumed stationary condition. Here we aim at analyzing simultaneous spike trains for time-dependent HOCs to trace active assemblies. We suggest to estimate the dynamics of HOCs by means of a state-space analysis with a log-linear observation model. A log-linear representation of the parallel spikes provides a well-defined measure of HOC based on information geometry (Amari, IEEE Trans. Inf. Theory (47) 1701-1711, 2001). We developed a nonlinear recursive filtering/smoothing algorithm for the time-varying log-linear model by applying a log-quadratic approximation to its posterior distribution. The time-scales of each parameter and their covariation are automatically optimized via the EM-algorithm under the maximum likelihood principle. To obtain the most predictive model, we compare the goodness-of-fit of hierarchical log-linear models with different order of interactions using the Akaike information criterion (AIC; Akaike, IEEE Trans. Autom. Control (19) 716-723, 1974). While inclusion of increasingly higher-order interaction terms improves model accuracy, estimation of higher-order parameters suffers from large variances due to the paucity of synchronous spikes in the data. This bias-variance trade-off is optimally resolved with the model that minimizes the AIC. Complexity of the model is thus selected based on the sample size of the data and the prominence of the higher-order structure. Application of the proposed method to simultaneous recordings of neuronal activity is expected to provide us with new insights into the dynamics of assembly activities, their composition, and behavioral relevance.
  • State-space analysis on time-varying higher-order spike correlations  [通常講演]
    Shimazaki H, Amari S, Brown EN, Grün S
    NIPS2008 Workshop on `Statistical Analysis and Modeling of Response Dependencies in Neural Populations' 2008年12月 口頭発表(招待・特別) Whistler, Canada
  • 多細胞同時記録スパイク時系列データの状態空間モデル  [通常講演]
    島崎 秀昭, ソーニャ グリューエン
    第11回情報論的学習理論ワークショップ 2008年10月 ポスター発表 仙台
  • 島崎 秀昭, 篠本 滋
    日本神経回路学会 第18回全国大会 2008年09月 ポスター発表 筑波 日本神経回路学会
     
    We propose a locally adaptive kernel method in estimating a spike-rate of a Poisson process. We select local bandwidths with which the estimated rate best fits to the underlying rate within local intervals. In addition, we optimize the local interval lengths so that the estimated rate fits to the underlying rate in an entire observation period. Numerical examples demonstrate that the proposed method performs better than the fixed kernel method and the classical Abramson’s adaptive kernel method with regard to the goodness-of-fit to the underlying rate.
  • Estimating time-dependent higher-order interactions in parallel spike trains  [通常講演]
    Shimazaki H, Gruen S
    Neuro2008 2008年07月 ポスター発表 Tokyo, Japan
  • Shimazaki H, Amari S, Brown EN, Grün S
    Statistical Analysis of Neuronal Data (SAND4) 2008年05月 ポスター発表 Pittsburgh, USA
  • Kernel width optimization in the spike-rate estimation  [通常講演]
    Shimazaki H, Shinomoto S
    Neural Coding 2007 2007年11月 ポスター発表 Montevideo, Uruguay
  • Optimization of a histogram of spike data  [通常講演]
    Shimazaki H, Shinomoto S
    Neuro2007 2007年09月 ポスター発表 Yokohama, Japan
  • ヒストグラムのビン幅の選択方法 -最適なPSTHの作り方  [招待講演]
    島崎 秀昭
    マルチニューロン研究会 2007年06月 公開講演,セミナー,チュートリアル,講習,講義等
  • A recipe for optimizing a time histogram of spike data  [通常講演]
    Shimazaki H
    RIKEN BSI Forums 2007年04月 Wako, Saitama, Japan RIKEN Brain Science Institute
  • A recipe for constructing a peri-stimulus time histogram  [通常講演]
    Shimazaki H
    The Boadian Seminar at Mind/Brain Institute 2007年03月 公開講演,セミナー,チュートリアル,講習,講義等 Baltimore, USA Johns Hopkins University
  • A recipe for optimizing a time-histogram with variable bin sizes  [通常講演]
    Shimazaki H, Shinomoto S
    Computational and Systems Neuroscience (Cosyne) 2007 2007年02月 ポスター発表 Salt Lake City, USA
  • A recipe for optimizing a time-histogram  [通常講演]
    Shimazaki H, Shinomoto S
    Neural Information Processing Systems (NIPS) 2006 2006年12月 ポスター発表 Whistler, Canada
  • スパイク時系列の潜時を補正して時間ヒストグラムを最適化する方法  [通常講演]
    島崎 秀昭
    日本神経回路学会第16回全国大会 2006年09月 口頭発表(一般)
  • Self-organized criticality by natural selection  [通常講演]
    Shimazaki H
    Frontiers in Dynamics: Physical and Biological Systems. 2006年05月 ポスター発表 Tokyo, Japan
  • Recipes for constructing an optimal time histogram  [通常講演]
    Shimazaki H
    Statistical Analysis of Neuronal Data (SAND3) 2006年05月 ポスター発表 Pittsburgh, USA
  • スパイク時系列ヒストグラムのビン幅の選択方法 -最適なPSTHの作り方  [通常講演]
    島崎 秀昭
    脳と心のメカニズム 第6回冬のワークショップ 2006年01月 留寿都,北海道
  • 乗法的競争モデルに見られる相転移  [通常講演]
    島崎 秀昭
    経済物理学 II -社会・経済への物理学的アプローチ- 2005年12月 ポスター発表 京都大学 湯川記念館
  • スパイク時系列のヒストグラム作成における最適区間幅決定のレシピ  [通常講演]
    島崎秀昭, 篠本滋
    日本神経回路学会第15回全国大会 2005年09月 ポスター発表 鹿児島
  • 離散力学系の競争モデルに見られる相転移と自然選択による転移点への接近  [通常講演]
    島崎 秀昭
    日本物理学会第60回年次大会 2005年03月 埼玉
  • Bose-einstein condensation in competitive processes  [通常講演]
    島崎 秀昭
    日本物理学会第59回年次大会 2004年03月 口頭発表(一般) 博多

その他活動・業績

  • Miguel Aguilera, Masanao Igarashi, Hideaki Shimazaki 2022年05月19日 
    Most systems in nature operate far from equilibrium, displaying time-asymmetric, irreversible dynamics. When a system's elements are numerous, characterizing its nonequilibrium states is challenging due to the expansion of its state space. Inspired by the success of the equilibrium Ising model in investigating disordered systems in the thermodynamic limit, we study the nonequilibrium thermodynamics of the asymmetric Sherrington-Kirkpatrick system as a prototypical model of large-scale nonequilibrium processes. We employ a path integral method to calculate a generating functional over the trajectories to derive exact solutions of the order parameters, conditional entropy of trajectories, and steady-state entropy production of infinitely large networks. The order parameters reveal order-disorder nonequilibrium phase transitions as found in equilibrium systems but no dynamics akin to the spin-glass phase. We find that the entropy production peaks at the phase transition, but it is more prominent outside the critical regime, especially for disordered phases with low entropy rates. While entropy production is becoming popular to characterize various complex systems, our results reveal that increased entropy production is linked with radically different scenarios, and combining multiple thermodynamic quantities yields a more precise picture of a system. These results contribute to an exact analytical theory for studying the thermodynamic properties of large-scale nonequilibrium systems and their phase transitions.
  • Hideaki Shimazaki arXiv 2006.13158 2020年06月23日 [査読無し][通常論文]
     
    This article reviews how organisms learn and recognize the world through the dynamics of neural networks from the perspective of Bayesian inference, and introduces a view on how such dynamics is described by the laws for the entropy of neural activity, a paradigm that we call thermodynamics of the Bayesian brain. The Bayesian brain hypothesis sees the stimulus-evoked activity of neurons as an act of constructing the Bayesian posterior distribution based on the generative model of the external world that an organism possesses. A closer look at the stimulus-evoked activity at early sensory cortices reveals that feedforward connections initially mediate the stimulus-response, which is later modulated by input from recurrent connections. Importantly, not the initial response, but the delayed modulation expresses animals' cognitive states such as awareness and attention regarding the stimulus. Using a simple generative model made of a spiking neural population, we reproduce the stimulus-evoked dynamics with the delayed feedback modulation as the process of the Bayesian inference that integrates the stimulus evidence and a prior knowledge with time-delay. We then introduce a thermodynamic view on this process based on the laws for the entropy of neural activity. This view elucidates that the process of the Bayesian inference works as the recently-proposed information-theoretic engine (neural engine, an analogue of a heat engine in thermodynamics), which allows us to quantify the perceptual capacity expressed in the delayed modulation in terms of entropy.
  • Shashwat Shukla, Hideaki Shimazaki, Udayan Ganguly 2019年11月14日 [査読無し][通常論文]
     
    The Bayesian view of the brain hypothesizes that the brain constructs a generative model of the world, and uses it to make inferences via Bayes' rule. Although many types of approximate inference schemes have been proposed for hierarchical Bayesian models of the brain, the questions of how these distinct inference procedures can be realized by hierarchical networks of spiking neurons remains largely unresolved. Based on a previously proposed multi-compartment neuron model in which dendrites perform logarithmic compression, and stochastic spiking winner-take-all (WTA) circuits in which firing probability of each neuron is normalized by activities of other neurons, here we construct Spiking Neural Networks that perform \emph{structured} mean-field variational inference and learning, on hierarchical directed probabilistic graphical models with discrete random variables. In these models, we do away with symmetric synaptic weights previously assumed for \emph{unstructured} mean-field variational inference by learning both the feedback and feedforward weights separately. The resulting online learning rules take the form of an error-modulated local Spike-Timing-Dependent Plasticity rule. Importantly, we consider two types of WTA circuits in which only one neuron is allowed to fire at a time (hard WTA) or neurons can fire independently (soft WTA), which makes neurons in these circuits operate in regimes of temporal and rate coding respectively. We show how the hard WTA circuits can be used to perform Gibbs sampling whereas the soft WTA circuits can be used to implement a message passing algorithm that computes the marginals approximately. Notably, a simple change in the amount of lateral inhibition realizes switching between the hard and soft WTA spiking regimes. Hence the proposed network provides a unified view of the two previously disparate modes of inference and coding by spiking neurons.
  • 中江健, 浦久保秀俊, 東広志, 田中康裕, 島崎秀昭, 尾藤晴彦, 石井信 日本神経回路学会誌 26 (3) 99 -103 2019年09月 [査読無し][通常論文]
  • 島崎秀昭, 小川正, 熊田孝恒 日本神経回路学会誌 26 (3) 49 -50 2019年09月 [査読無し][通常論文]
  • Hideaki Shimazaki arXiv 1902.11233 2019年02月 [査読無し][通常論文]
     
    How do organisms recognize their environment by acquiring knowledge about the world, and what actions do they take based on this knowledge? This article examines hypotheses about organisms' adaptation to the environment from machine learning, information-theoretic, and thermodynamic perspectives. We start with constructing a hierarchical model of the world as an internal model in the brain, and review standard machine learning methods to infer causes by approximately learning the model under the maximum likelihood principle. This in turn provides an overview of the free energy principle for an organism, a hypothesis to explain perception and action from the principle of least surprise. Treating this statistical learning as communication between the world and brain, learning is interpreted as a process to maximize information about the world. We investigate how the classical theories of perception such as the infomax principle relates to learning the hierarchical model. We then present an approach to the recognition and learning based on thermodynamics, showing that adaptation by causal learning results in the second law of thermodynamics whereas inference dynamics that fuses observation with prior knowledge forms a thermodynamic process. These provide a unified view on the adaptation of organisms to the environment.
  • Safura Rashid Shomali, Majid Nili Ahmadabadi, Seyyed Nader Rasuli, Hideaki Shimazaki bioRxiv 10.1101/479956 2018年11月 [査読無し][通常論文]
     
    bioRxiv. (2018) doi:
  • 島崎 秀昭, 吉田 正俊, 田口 茂, 磯村 拓哉, 田中 琢真, 大羽 成征, 乾 敏郎 日本神経回路学会誌 25 (3) 51 -52 2018年09月 [査読無し][通常論文]
  • 島崎 秀昭, 大羽 成征, 吉田 正俊 日本神経回路学会誌 24 (4) 228 -232 2017年12月 [査読無し][招待有り]
  • MaBouDi H, Shimazaki H, Soltanian-Zadeh H, Amari S bioRxiv 2017年03月 [査読無し][通常論文]
  • Hideaki Shimazaki, Kolia Sadeghi, Tomoe Ishikawa, Yuji Ikegaya, Taro Toyoizumi SCIENTIFIC REPORTS 5 2015年10月 [査読無し][通常論文]
  • 島崎 秀昭 脳科学若手の会 2014年01月25日 [査読無し][招待有り]
  • Fukai T, Shimazaki Optogenetic activation of an inhibitory network enhances feedforward functional connectivity in auditory cortex 2014年01月 [査読無し][招待有り]
     
    Fukai T and Shimazaki H: F1000Prime Recommendation of [Hamilton LS et al., Neuron 2013,80(4):1066-76]. Faculty of 1000, 10 Jan 2014; DOI: 10.3410/f.718185775.793489321.
  • 島崎 秀昭 生物物理 53 (2) 103 -104 2013年 [査読有り][招待有り]
  • 島崎 秀昭 日本神経回路学会誌 18 (4) 194 -203 2011年12月 [査読有り][招待有り]
     
    神経細胞集団は協調して情報処理を行っています.特に比較的短い時間内に複数の細胞が発火する同期発火現象はセル·アセンブリ仮説やシンファイア·チェーンといった細胞集団による情報処理の仮説との関係から注目を集めてきました.近年実データの同期的発火構造を調べるために対数線形モデルを用いた解析が広く行われています.しかしこれまでの対数線形モデルは定常性を仮定しており行動下の動物から得られるスパイクデータへ適用することはできませんでした.本稿は対数線形モデルの基本事項,パラメータの基本的な推定法(最尤推定·MAP推定)を経て,最終的に動的な相関構造の推定を可能にする非線形フィルタの構築に至るまでを紹介します.
  • Shigeru Shinomoto, Hideaki Shimazaki, Takeaki Shimokawa NEUROSCIENCE RESEARCH 68 E50 -E51 2010年 [査読無し][通常論文]
  • Hideaki Shimazaki, Sonja Gruen NEUROSCIENCE RESEARCH 61 S140 -S140 2008年 [査読無し][通常論文]
  • 島崎秀昭 日本神経回路学会誌 = The Brain & neural networks 13 (2) 84 -85 2006年06月 [査読無し][通常論文]
  • 小林祐喜, 島崎秀昭, 荻久保佳伸, 溝口健二, 相原威, 塚田稔 電子情報通信学会技術研究報告. NC, ニューロコンピューティング 99 (686) 31 -36 2000年03月15日 [査読無し][通常論文]
     
    近年、軸索を逆伝搬する発火活動が報告されており、海馬における学習則を考える場合、入力同士の同時性以外に入力と軸索を逆伝搬する発火活動(出力逆伝搬)の同時性も考慮する必要がある。そこで本研究は、電気刺激をCA1野へのびるshaffer側枝と出力層であるstratum oriensに刺入し、その入力と出力逆伝搬の位相をτ=0ms, ±5ms, ±10ms, ±20ms, ±25mS, ±50ms, ±100ms, ±500ms, 1000msと変化させたときのLTP(長期増強), LTD(長期抑圧)の空間分布を光計測法を用いて計測した。結果としてLTP, LTDは位相差に依存して空間的に異なった分布として引き起こされることが明らかとなった。
  • 島崎秀昭, 小林祐喜, 大澤さや香, 森田稔, 溝口健二, 相原威, 塚田稔 電子情報通信学会技術研究報告. NC, ニューロコンピューティング 99 (686) 37 -43 2000年03月15日 [査読無し][通常論文]
     
    シナプス入力と樹状突起上の逆伝播発火がシナプス結合に到達するタイミング(発火タイミング)がシナプス可塑性に果たす役割が注目されている。本研究では、この問題に対し光計測法によるアプローチを提案する。シナプス入力と逆伝播発火を独立に計測し、同時刺激時の発火タイミングを空間的に推定した。これを対応するLTP・LTDの空間分布と重ね合わせ、シナプス可塑性の発火タイミング依存性を示す特性曲線を求めた。光計測法によるアプローチではガラス電極法では不可能な試行数を一度に行なえるだけでなく、特性曲線の空間的な特異性も容易に検討できる。その結果、発火タイミング+20ms付近にLTDを確認し、これが抑制性のインターニューロンに由来することを示唆する特性曲線の空間特異性も確認した。

特許

  • 特願2020-007811:情報処理装置、情報処理方法およびプログラム  2020年01月21日
    (国)京都大学/HRIホンダ

受賞

  • 2016年10月 The 23rd International Conference on Neural Information Processing Excellent paper award
     
    受賞者: Christian Donner;Hideaki Shimazaki
  • 2009年09月 日本神経回路学会大会研究賞
     
    受賞者: 島崎 秀昭
  • 2006年09月 日本神経回路学会奨励賞
     
    受賞者: 島崎 秀昭

共同研究・競争的資金等の研究課題

  • 大規模・非線形な神経細胞集団活動を可視化する統計解析技術の開発
    日本学術振興会:科学研究費助成事業 基盤研究(C)
    研究期間 : 2020年04月 -2024年03月 
    代表者 : 島崎 秀昭
  • 2021年度 分野融合型共同研究事業 共同研究型
    研究期間 : 2021年03月 -2022年04月 
    代表者 : 吉田 正俊 , 磯村 拓哉 , Christopher Buckley , 本田 直樹, 磯田 昌岐
  • サイボーグAI学習のための階層的ベイズネットワーク
    国立研究開発法人新エネルギー・産業技術総合開発機構(NEDO):人と共に進化する次世代人工知能に関する技術開発事業
    研究期間 : 2020年09月 -2022年03月
  • 課題名なし
    日本神経回路学会:日本神経回路学会30周年記念研究助成金
    研究期間 : 2021年02月 -2022年02月
  • 自然科学研究機構:2020年度 分野融合型共同研究事業 共同研究型
    研究期間 : 2020年04月 -2021年03月 
    代表者 : 吉田 正俊, 磯村 拓哉, Christopher Buckley, Shashwat Shukula, 磯田 昌岐
  • 認知神経科学の先端 脳の理論から身体・世界へ
    自然科学研究機構生理学研究所:2019年度自然科学研究機構生理学研究所 共同利用研究
    研究期間 : 2019年04月 -2020年03月 
    代表者 : 島崎 秀昭
  • 脳の自由エネルギー原理 チュートリアル・ワークショップ
    自然科学研究機構:2019年度 自然科学研究機構 分野融合型共同研究事業 ワークショップ
    研究期間 : 2019年03月 -2019年04月 
    代表者 : 島崎 秀昭
  • 日本学術振興会特別研究員
    日本学術振興会:特別研究員制度
    研究期間 : 2008年 -2012年 
    代表者 : 島崎 秀昭
     
    研究課題番号:08J01386 数物系科学 統計科学 神経科学への応用を目的とした高次相関を持つスパイク時系列群の作成と推定手法の開発
  • 優秀若手研究者海外派遣事業 (特別研究員)
    日本学術振興会:優秀若手研究者海外派遣事業
    研究期間 : 2010年 -2011年 
    代表者 : 島崎 秀昭
  • 神経科学への応用を目的とした高次相関を持つスパイク時系列群の作成と推定手法の開発
    日本学術振興会:科学研究費助成事業 特別研究員奨励費
    研究期間 : 2008年 -2010年 
    代表者 : 島崎 秀昭
     
    平成22年度は,本研究で提案するスパイク相関の状態空間モデルに対して前年度に構築した2つの評価手法である(i)周辺尤度の近似解を用いた情報量基準によるモデル選択法,(ii)周辺尤度比(ベイズ因子)を用いた相関モデルの検定法,についてシミュレーションによる評価を行ないこれらの有用性を検証した.前者では実際に汎化誤差を数値計算により求め,複数の情報量基準と比較して赤池による情報量基準がもっとも妥当であることを確認した.後者では同期スパイクを生成する背後のスパイク相関モデルについて,周辺尤度比とサロゲート法と組み合わせることでデータからその存在を検定する枠組みを構築した.本手法では潜在変数であるスパイク相関の構造として対立する二つのモデル群を考え,観測スパイクデータに対する各々の周辺尤度の比を算出する.周辺尤度比の有意水準を決定し仮説検定を行うため,対象モデルの相関が存在しないという帰無仮説の下でリサンプリングされたサロゲートデータに対して周辺尤度比を算出した.サロゲートデータの作成は推定された状態空間モデルを低次元に射影したモデルを用いることで実現した.本手法により動物の行動に応じて現れると考えられるスパイク相関構造の検定を行うことが可能となった.一連の研究により,神経スパイクデータから細胞集団の動的な協調活動を検出し,動物の認知・行動に果たす役割を明らかにするための解析技術が実用段階に達したと考える.
  • 日本学術振興会特別研究員
    日本学術振興会:特別研究員制度
    研究期間 : 2006年 -2008年 
    代表者 : 島崎 秀昭
     
    研究課題番号:06J02651 数物系科学 統計科学 神経科学への応用を目的としたポイントプロセスからの強度過程の統計的推定問題
  • 神経科学への応用を目的としたポイントプロセスからの強度過程の統計的推定問題
    日本学術振興会:科学研究費助成事業 特別研究員奨励費
    研究期間 : 2006年 -2007年 
    代表者 : 島崎 秀昭
     
    電気神経生理学の動物実験では感覚刺激・行動・注意等と神経細胞の発火頻度(レート)の相関関係がよく調べられます.広く使われているレート推定の手法に,同一刺激下で行われた複数回の試行のスパイク時系列を適当な時間幅をもつ区間に分割し,その中でのスパイク生成数を棒グラフとして表すPeristimulus Time Histogram(PSTH)があります. PSTHの形状は分割する区間の時間幅(ビン幅)に依存するにもかかわらず,区間幅は多くの場合研究者により恣意的に決められてきました,そこで我々は実験データから最適区間幅を導出する手順を考案しました.また,実験データ数が非常に少ない場合は最適幅が非常に大きくなってしまうことがあります.こうした場合に用いる指針として,我々は所望の区間幅を得るのに必要な実験回数を推定する方法も考案しました. これらの研究成果は神経科学データの解析だけでなく,一般の密度推定においても活用できます.

教育活動情報

主要な担当授業

  • 大学院共通授業科目(教育プログラム):人間知・脳・AI教育プログラム
    開講年度 : 2021年
    課程区分 : 修士課程
    開講学部 : 大学院共通科目
    キーワード : 人文科学 社会科学 神経科学 人工知能 脳 人間 機械学習 統計学
  • 大学院共通授業科目(教育プログラム):人間知・脳・AI教育プログラム
    開講年度 : 2021年
    課程区分 : 修士課程
    開講学部 : 大学院共通科目
    キーワード : 人文科学、社会科学、神経科学、人工知能、脳、神経科学、人間、機械学習、統計モデリング
  • 大学院共通授業科目(教育プログラム):人間知・脳・AI教育プログラム大学院共通授業科目(教育プログラム
    開講年度 : 2021年
    課程区分 : 修士課程
    開講学部 : 大学院共通科目
    キーワード : 人文科学、社会科学、神経科学、人工知能、脳、人間、機械学習、統計モデリング

大学運営

委員歴

  • 2021年01月 - 2022年07月   日本神経科学会   第45回日本神経科学大会プログラム委員
  • 2018年04月 - 2020年03月   日本神経回路学会   理事
  • 2014年02月 - 2016年03月   文部科学省科学技術政策研究所科学技術動向研究センター   専門調査員


Copyright © MEDIA FUSION Co.,Ltd. All rights reserved.