[23a-61C-1~3] 【Open Symposium】Informatics for Materials Science and Brain Science

9:50 AM - 10:40 AM JST(12:50 AM - 1:40 AM UTC)

[23a-61C-2] Bayesian mechanics of classical, neural, and quantum systems

〇Takuya Isomura¹ (1.RIKEN CBS)

Keywords:

free energy principle,neural networks,Bayesian inference

Bayesian mechanics is a field for conceptualising dynamical systems as Bayesian inferences. In this talk, we will introduce that the Hamiltonian of a general dynamical system corresponds to a class of generative models, so that the Helmholtz energy of the system is equivalent to variational free energy under the identified generative model. We will show that this property naturally emerges in real systems using examples of coupled oscillators, neural network models, in vitro neural networks, and quantum computers.