Extending the Bayesian Framework from Information to Action

TL;DR

This paper explores an extended Bayesian inference framework that combines standard and inverse Bayesian methods, highlighting its implications for biological information processing and proposing new experimental setups.

Contribution

It introduces a dual Bayesian inference approach that enriches probability spaces and elucidates differences between biological and artificial information processing.

Findings

Dual Bayesian inference facilitates discovery.

Biological processing utilizes nonlinearities unlike artificial systems.

Proposes experimental setup to test extended Bayesian ideas.

Abstract

In this review, we examine an extended Bayesian inference method and its relation to biological information processing. We discuss the idea of combining two modes of Bayesian inference. The first is the standard Bayesian inference, which contracts probability space. The second is its inverse, which extends and enriches the probability space of latent and observable variables. Their combination has been observed that, greatly, facilitates discovery. Moreover, this dual search during the updating process elucidates a crucial difference between biological and artificial information processing. The latter is restricted due to nonlinearities, while the former utilizes it. This duality is ubiquitous in biological information process dynamics (`flee-or-fight', `explore-or-exploit' etc.) as is the role of fractality and chaos in its underlying nonequilibrium, nonlinear dynamics. We also propose a new experimental set up that stems from testing these ideas.