第14回 冬のワークショップ

脳の計算論の未来 "Future perspectives of computational brain science"

日程:2014年1月8日(水)-10日(金)

会場:
ルスツリゾート(北海道蛇田郡留寿都村字泉川13)
  http://www.rusutsu.co.jp/winter/
会場地図:
ノースウイング コンベンションホール 18番ホール 

スケジュール:

 1月8日(水)-10日(金)
脳の計算論の未来 "Future perspectives of computational brain science"
8日 スペシャルセッション
18:10-19:00 甘利俊一(RIKEN, BSI)
19:10-20:00 Erik De Schutter(OIST)
20:10-21:00 川人光男(ATR脳情報研究所)
21:00-23:00 ポスターセッション
9日 トピックセッション
15:30-16:20 田中宏和 (北陸先端科学技術大学院大学)
16:30-17:20 山田真希子 (放射線医学総合研究所)
17:30-18:20 西本伸志(情報通信研究機構)
20:00-22:00 ポスターセッション
10日 トピックセッション
 9:00-9:50 Joshua Johansen (RIKEN,BSI)
 10:00-10:50 松本正幸(筑波大学)
 11:00-11:50 中原裕之(RIKEN,BSI)




Abstracts and References:

Shun-ichi Amari
甘利俊一 
理化学研究所脳科学総合研究センター RIKEN Brain Science Institute


Dreaming of Mathematical Neuroscience for Half a Century

Science begins with observation of phenomena. They are then categorized for systematic descriptions and consolidated as knowledge. The final stage is to search for the principles underlying the phenomena. Brain science proceeds along this line.
However, one might argue that brain science is different from physical science. Its phenomena are multiple, from the molecular level, cellular level, systems level to mental level. Moreover, the brain has emerged via a long history of evolution, different from simple physical world. This might make it very difficult, even impossible, to establish mathematical neuroscience.
This is true. But I believe that there are a number of principles in information processing in the brain. The structure which fits the principles was found via evolution and has fixed in the brain, although their realizations are very complex because of this historical reason.
Mathematical neuroscience searches for the principles by using very simple abstract or unrealistic models. If we can find principles by using such simple models, we then proceed to construct more realistic models in which the principles are realized in very complex forms like the real brain. This latter is the role of computational neuroscience.
I have been dreaming of such an approach for half a century and still searching for it. I will talk some fundamental problems concerning neurodynamics, and learning and self-organization, hoping the establishment of mathematical neuroscience.

References:
Neural Networks"Dreaming of mathematical neuroscience for half a century"



Erik De Schutter 
Okinawa Institute of Science and Technology


Challenges in multi-scale modeling of the brain

It is generally accepted that brain function requires the constant interaction between different spatial levels of organization: from the genetic and molecular level, over cellular and network levels, all the way up to large brain structures. Most modeling studies of nervous function have been, however, restricted to a single spatial level and either ignore other levels or treat them as random factors. Therefore it is not surprising that there is now a growing interest in multi-scale modeling. Most success has been achieved till now in combining the cellular and network levels, as is for example done in the simulations of a cortical column performed by the Blue Brain Project.
A major initiative to promote multi-scale simulation is the (European) Human Brain Project which will develop extensive software support to run multi-scale models. I will briefly describe the these goals of the Human Brain Project, which focus mostly on simulator development and high performance computing. I will also show how we approach multi-scale simulation in the STEPS simulator that we have been developing. Next I will discuss two challenges which have received much less attention: flexible, simulator-independent model description and between level model transformation.


Mitsuo Kawato
川人光男 
ATR脳情報研究所


Computational-Model Based Decoded Neurofeedback

Computational-model based neuroimaging and computational-model based neruophysiology are successful research paradigms that directly investigate experimental plausibility of computational models. Typically, computational variables are estimated from behavioral data consisting of stimuli, actions and rewards on trial by trial basis and they are used as explanatory variables to regress fMRI BOLD signal or neural firings. One big drawback of this approach is that it is correlational rather than causal. Neurofeedback techniques especially fMRI decoded neurofeedback could provide an entirely different research paradigm examing cause (brain activity) and effect (behavior, mind). However, computational models did not explicity contribute to this. Here, I propose a new apporach combining these two. I explain research agenda to tackle visual consciousness based on computational-model based decoded neurofeedback.


Hirokazu Tanaka
田中宏和
北陸先端科学技術大学院大学 Japan Advanced Institute of Science and Technology


第一次運動野は空間ベクトルを用いて到達運動ダイナミクスを計算する / Computing Reaching Dynamics in Motor Cortex with Cartesian Spatial Coordinates

身体との体部位対応を持つ第一次運動野(M1)は、錐体路繊維を通して脊髄中の運動ニューロンに投射するが、ど のようにして身体運動を制御す るかについては完全には理解されていない。ダイナミクスの立場ではM1は身体 座標系における筋張力や関節トルクを制御すると主張するが、キネ マティクスの立場では外部座標系における 運動軌道を表現しているとする。本講演では、外部座標系を用いると到達運動を記述する運動方程式が単 純に なることを示し、開n-リンク系の一般式を導出する (ニュートン-オイラー方程式)。この式は左辺のダイナミカ ル量である関節トルクと右辺のキネマティカル量である空間運動を直接関係づける式であり、関節 角表現を明 示的に必要としないため、逆キネマティクス計算や不良設定性問題を回避することができる。ベクトル外積を用 いて理想軌道を到達運動 ダイナミクスへと変換しているという計算理論を提案する。ベクトル外積各項をM1神 経細胞の発火頻度と同一視すれば、(1)コサイン・チュー ニング、(2)最適方位の作業空間依存性、(3) 複数座 標系の混在、(4)運動適応後の活動変化、 (5)ポピュレーションベクトルの時空間的性質といった実験結果を説 明できる。筋張力の計算も一層のニューラルネットワークで計算できることを示す。これ らの結果から、キネ マティクスとダイナミクスの立場が相反するものではなく、運動方程式における左辺と右辺を強調しているにす ぎないと結論す る。最後に、この計算モデルを実験的に検証する試みに関して簡単に触れる。

References:
Hirokazu Tanaka and Terrence J. Sejnowski J Neurophysiol 109: 1182–1201, 2013


Makiko Yamada
山田真希子 
放射線医学総合研究所 / National Institute of Radiological Sciences


認知バイアスの意識体験と脳内分子メカニズム / Molecular mechanisms of cognitive bias and the confidence

私たちの知覚や思考は、歪みやバイアスがかかっている。例えば、私たちは自分のことを平均より優れていると考え、悪い出来事は自分の身には起こらないと信じ、成功は自分の活躍の結果だとみなす。これらは「ポジティブ錯覚」と呼ばれる認知バイアスであり、私たちは概して自分に都合良く物事を捉えている。非現実的であったとしても、ポジティブ錯覚を持つことで、人は自分の可能性を信じて目標に向かうことができる。また、知覚や思考の錯覚は、例え真実を知らされたとしても、修正されにくいという特徴を持つ。つまり、錯覚に対する確信はそれほど確固とした意識体験なのである。この自己の信念に対する信念は「メタ認知」と呼ばれ、錯覚や幻覚・妄想などの形成と維持に関与している。認知バイアスとメタ認知の脳機能とその背後にある分子機能を解明することで、私たちの信じる現実、信念、そして意識体験が脳内でどのように生成されているかを知る手がかりとなる。講演では、fMRI(脳機能画像)とPET(陽電子放射断層撮像)を組み合わせた最新の研究結果を紹介する。

References:
Superiority illusion arises from resting-state brain networks modulated by dopamine


Shinji Nishimoto
西本伸志 
情報通信研究機構 NICT CiNet / 大阪大学 Osaka University


モデリングアプローチ:脳神経活動の定量的理解を目指す新たな枠組み / Modeling approach: a new framework for quantitative understanding of brain activity

我々の日常を構成する自然な体験は、複雑で多様、かつダイナミックである。 脳神経科学のゴールの一つは、このような自然で複雑な知覚・認知体験を司る脳 機能を解明することにある。本講演では、自然条件下における脳機能の定量的理 解を目指すための新たな枠組みとして、モデリングアプローチを紹介したい。こ のアプローチでは、脳神経情報処理に関する仮説は任意の自然条件下における脳 活動を予測する定量的モデル(エンコーディングモデル)として実装され、仮説 の妥当性は新規条件下における活動予測性能によって評価される。モデリングア プローチは汎用的なものであり、その適用例は初期視覚野における時空間情報表 現から高次領野における意味情報表現まで、受動的知覚条件下から能動的認知タ スク条件下まで、単一細胞電位記録から機能的磁気共鳴画像(fMRI)記録まで、 多岐に渡る。更に、 エンコーディングモデルは脳活動から体験を読み出すデコ ーディングを行うための基盤としても用いる事ができる。モデリングアプローチ は、自然で複雑な体験を司る脳活動の理解を促進し、予測に基づいた定量的神経 科学を推進するための強力な手段を提供する。

References:
Çukur T, Nishimoto S, Huth AG, Gallant JL (2013) Attention during natural vision warps semantic representation across the human brain.
Nature Neuroscience, 16(6):763-770.

Huth AG, Nishimoto S, Vu AT, Gallant JL. (2012) A continuous semantic space describes the representation of thousands of object and action categories across the human brain.
Neuron, 76(6):1210-1224.

Nishimoto S, Vu AT, Naselaris T, Benjamini Y, Yu B, Gallant JL. (2011) Reconstructing visual experiences from brain activity evoked by natural movies.
Current Biology, 21(19):1641-1646.



Joshua Johansen  
RIKEN Brain Science Institute


A neural circuit mechanism for triggering and setting the strength of fear memories

Aversive experiences are powerful triggers for neural plasticity and memory formation and the intensity of these experiences controls the strength of the memory. To trigger memories, aversive experiences activate neural ‘teaching signal’ circuits which engage plasticity in brain regions involved in learning and memory. Fear conditioning is an ideal model system for studying these processes because a site of plasticity mediating memory formation has been identified in the lateral nucleus of the amygdala. Using a combined optogenetic, behavioral and physiological approach, we examined the factors and neural circuits which trigger fear learning, how aversive teaching signals are encoded within the fear teaching signal circuit and the functional implications of this neural coding for setting the strength of fear memories. The results suggest a concerted neural mechanism for how aversive experiences initiate and control the strength of associative fear memories and have important implications for our understanding of anxiety disorders characterized by exaggerated aversive learning.


Masayuki Matsumoto
松本正幸 
筑波大学医学医療系 Faculty of Medicine, University of Tsukuba


ドーパミンニューロンは何をコードしているのか?/ What signals do dopamine neurons encode?

Dopamine neurons in the substantia nigra pars compacta and the ventral tegmental area are well known for their phasic responses to rewards and cues that predict reward. However, in contrast to their accepted role in reward processing, there has been considerable debate over the role of dopamine neurons in processing non-rewarding events. Some theories suggest that dopamine neurons primarily signal rewarding events, while others suggest that they encode additional signals related to surprising, novel, salient, and even aversive experiences. Supporting the latter theories, recent studies reported that, although a group of dopamine neurons was inhibited by aversive events as they encoded the value-related signal, another group of dopamine neurons was excited. Since the neurons with excitatory responses to aversive events were excited by rewarding events as well, they were presumed to encode motivational salience rather than motivational value. Based on these findings, it is proposed that dopamine neurons are not a homogeneous population and are divided into multiple groups encoding distinct signals suitable for different functions. In this talk, I will present our recent studies supporting this theory.

References:
Matsumoto M, Takada M, Distinct representations of cognitive and motivational signals in midbrain dopamine neurons.
Neuron, 79, p1011-24, 2013

Matsumoto M, Hikosaka O, Two types of dopamine neuron distinctly convey positive and negative motivational signals.
Nature, 459, p837-41, 2009



Hiroyuki Nakahara
中原裕之 
理化学研究所 脳科学総合研究センター RIKEN Brain Science Institute


Toward a computational theory of mind: decision-making and neural coding

A fundamental challenge in social cognition is how one learns and predicts the mind of others; and what the underlying neural mechanisms are. Answering these questions is a challenge we approach by mapping behavioral-level complexity and related neural systems through computational frameworks, i.e., key computations. Extending reinforcement learning theory into the realm of social cognition and combining human fMRI with modeling, we recently addressed how one learns to simulate decision-making of others. Two learning signals were found in a hierarchical arrangement: (1) a reward prediction error, generated by simulating others’ valuation process by direct recruitment of one’s own process and encoded in the ventromedial prefrontal cortex, and (2) an action prediction error, generated by combining this simulation with observation of the other’s choices and encoded in the dorsolateral/dorsomedial prefrontal cortex. These findings show that this simulation employs a core prefrontal circuit for modeling the others’ valuation to generate a prediction and an adjunct circuit for tracking behavioral variations to refine the prediction. Toward a “computational theory of mind” in decision-making, I will outline our ongoing research and discuss how our pursuit is supported by basic research on decision-making and neural coding and they together will open up new avenues of inquiry in the future.

References:

理研・脳総研・理論統合脳科学研究チーム(こちらから、その他の論文も含めdownload可)。

Suzuki S, Harasawa H, Ueno K, Gardner JL, Ichinohe N, Haruno M, Cheng K, Nakahara H. (2012) Learning to simulate others’ decisions.
Neuron.74: 1125-1137.

Nakahara H, Hikosaka O. (2012) Learning to represent reward structure: A key to adapting to complex environments.
Neuroscience Research.74(3-4): 177-183.