In vitro neural networks minimise variational free energy

Abstract

In this work, we address the neuronal encoding problem from a Bayesian perspective. Specifically, we ask whether neuronal responses in an in vitro neuronal network are consistent with ideal Bayesian observer responses under the free energy principle. In brief, we stimulated an in vitro cortical cell culture with stimulus trains that had a known statistical structure. We then asked whether recorded neuronal responses were consistent with variational message passing (i.e., belief propagation) based upon free energy minimisation (i.e., evidence maximisation). Effectively, this required us to solve two problems: first, we had to formulate the Bayes-optimal encoding of the causes or sources of sensory stimulation, and then show that these idealised responses could account for observed electrophysiological responses. We describe a simulation of an optimal neural network (i.e., the ideal Bayesian neural code) and then consider the mapping from idealised in silico responses to recorded in vitro responses. Our objective was to find evidence for functional specialisation and segregation in the in vitro neural network that reproduced in silico learning via free energy minimisation. Finally, we combined the in vitro and in silico results to characterise learning in terms of trajectories in a variational information plane of accuracy and complexity.

What might interoceptive inference reveal about consciousness?

“…On this view, self-organizing systems appear to simulate the consequences of actions in order to select those actions that lead to the least free energy in the future (i.e., least action), leading to a balance between exploratory, information-seeking behaviour, and exploitative, pragmatic behaviour. Active inference (often modelled using Partially Observable Markov Decision Processes) has been used to describe a variety of phenomena, ranging from stratospheric adaption [105] through cellular organization [106], interoceptive processes [64], and neuronal activity [107,108] to psychiatric disorders [109,110].…”

Reverse engineering neural networks to characterise their cost functions

“…We 449 considered a BSS comprising two hidden sources (or states) and 32 observations (or sensory 450 inputs), formulated as an MDP. The two hidden sources comprised four patterns: $ = 451 $ (() ⨂ $ (D) = (0,0), (1,0), (0,1), (1,1 First, as in (Isomura & Friston, 2018), we demonstrated that a network with a cost function 461 with optimised constants (? 3( , 3B A = (− ln 2 , − ln 2) and ?…”

“…sources (Isomura et al, 2015). Furthermore, we showed that minimising variational free 600 energy under an MDP is sufficient to reproduce the learning observed in an in vitro network 601 (Isomura & Friston, 2018). Our framework for identifying biologically plausible cost functions 602 could be relevant for identifying the principles that underlie learning or adaptation processes 603 in biological neuronal networks, using empirical response data.…”

On the Achievability of Blind Source Separation for High-Dimensional Nonlinear Source Mixtures

“…Thus, it follows that major linear BSS algorithms for PCA (Oja, 1982(Oja, , 1989 and ICA (Bell & Sejnowski, 1995, 1997Amari et al, 1996;Hyvarinen & Oja, 1997) are formulated as a variant of Hebbian plasticity rules. Moreover, in vitro neuronal networks learn to separately represent independent hidden sources in (and only in) the presence of Hebbian plasticity (Isomura, Kotani, & Jimbo, 2015;Isomura & Friston, 2018). Nonetheless, the manner in which the brain can possibly solve a nonlinear BSS problem remains unclear, even though it might be a prerequisite for many of its cognitive processes such as visual recognition (DiCarlo et al, 2012).…”