Improving the Minimum Free Energy Principle to the Maximum Information Efficiency Principle

Area of Science:

• Neuroscience
• Information Theory
• Computational Biology

Background:

• The Minimum Free Energy Principle (FEP) by Friston, based on Variational Bayesian (VB) methods, suggests the brain and behavior self-organize with the environment.
• FEP has theoretical limitations, including potential misunderstandings and constraints limited to likelihood functions.

Purpose of the Study:

• To address FEP's limitations by introducing Semantic Variational Bayesian (SVB) and the Maximum Information Efficiency (MIE) principle.
• To provide a more comprehensive framework for understanding brain-environment interactions and active inference.

Main Methods:

• Introduction of the semantic information G theory and the R(G) function, utilizing a P-T probability framework.
• Application of logical Bayesian Inference and analysis of the R(G) function.
• Theoretic analysis and computing experiments to validate the proposed SVB and MIE principles.

Main Results:

• Demonstration that R - G = F - H(X|Y), where F is Variational Free Energy (VFE) and H(X|Y) is Shannon conditional entropy, challenging the notion of continuously decreasing F.
• SVB is shown to be a reliable and straightforward method for latent variable analysis and active inference.
• Clarification of the relationships between Shannon information, semantic information, VFE, free energy, exergy, and conditional entropy in different thermodynamic systems.

Conclusions:

• SVB and MIE offer advancements over FEP, providing a more nuanced understanding of information processing in biological systems.
• The proposed principles enhance the interpretability and applicability of free energy principles.
• Integration with deep learning methods is suggested for broader applications of the MIE principle.