[終了しました] ipi seminar [ハイブリッド開催] 2024年6月5日(水)10:30~12:00

知の物理学研究センター / Institute for Physics of Intelligence (iπ)


【日時・場所/Date・Venue】
2024年6月5日(水)10時30分~12時 /June 5, 10:30 - 12:00 (JST) @理学部1号館913号室&Zoom

【発表者/Speaker】
磯村 拓哉 氏(理化学研究所)

【タイトル/Title】
"Bayesian mechanics of self-organizing and evolutionary systems"

【概要/Abstract】
Bayesian mechanics is a framework that addresses dynamical systems that can be conceptualized as Bayesian inference. However, the elucidation of requisite generative models is required for empirical applications to realistic self-organizing systems. This talk introduces that the Hamiltonian of generic dynamical systems constitutes a class of generative models, thus rendering their Helmholtz energy naturally equivalent to variational free energy under the identified generative model. The self-organization that minimizes the Helmholtz energy entails matching the system's Hamiltonian with that of the environment, leading to an ensuing emergence of their generalized synchrony. In short, these self-organizing systems can be read as performing variational Bayesian inference of the interacting environment. These properties have been demonstrated with coupled oscillators, simulated and living neural networks, and quantum computers. Furthermore, it is shown that the ensemble Helmholtz energy minimization can derive Darwinian evolutionary law and active Bayesian model selection. These notions offer foundational characterizations and predictions regarding asymptotic properties of self-organizing systems exchanging with the environment, providing insights into potential mechanisms underlying emergence of intelligence.
References
1.Isomura, Bayesian mechanics of self-organising systems, arXiv:2311.10216 (2023), arXiv: 2311.10216 
2.Isomura, K. Kotani, Y. Jimbo, and K. J. Friston, Experimental validation of the free-energy principle with in vitro neural networks, Nature Communications 14, 4547 (2023), doi: 10.1038/s41467-023-40141-z 
3.Isomura, H. Shimazaki, and K. J. Friston, Canonical neural networks perform active inference, Communications Biology 5, 55 (2022), doi: 10.1038/s42003-021-02994-2 

これらの講演に関するZoomのリンク等の案内を受け取ることを希望されるかたは、下記のgoogle formからメールアドレスをご記入ください。こちらに登録頂いた情報は、案内の配信のみに利用いたします。

登録フォーム:https://forms.gle/xnLmd9Kc1BaaNPgq8

過去の発表リスト:https://www.phys.s.u-tokyo.ac.jp/about/17106/

          世話人:知の物理学研究センター 髙橋昂

⇦Top page

  • このエントリーをはてなブックマークに追加