Abstract
Biehl et al. (2021) present some interesting observations on an early formulation of the free energy principle. We use these observations to scaffold a discussion of the technical arguments that underwrite the free energy principle. This discussion focuses on solenoidal coupling between various (subsets of) states in sparsely coupled systems that possess a Markov blanket—and the distinction between exact and approximate Bayesian inference, implied by the ensuing Bayesian mechanics.
Highlights
We first rehearse the major steps in deriving the free energy principle (FEP) and focus on three cardinal issues addressed in Biehl et al.; namely, what is the precise form of the dynamical coupling among states that constitute a Markov blanket partition? What implications attend a nonzero evidence bound, when interpreting self-organisation as self-evidencing (i.e., Bayesian inference)? Further, when do variational free energy gradients vanish? The first of the three issues appears in Biehl et al
The flows can be fine-tuned to create Markov blankets in the absence of any sparsity constraints on coupling; the FEP only applies to Markov blankets that emerge under sparse flows; in particular, when autonomous states are uncoupled from external states
We looked at three fundamental issues that underpin the free energy principle
Summary
We use the observations of Biehl et al (ibid) to drill down on the interesting points they raise—and their implications in the setting of the FEP. To contextualise these observations, we first rehearse the major steps in deriving the FEP and focus on three cardinal issues addressed in Biehl et al.; namely, what is the precise form of the dynamical coupling among (subsets of) states that constitute a Markov blanket partition? One could read Biehl et al as a critique of early formulations of the FEP—in terms of implicit assumptions and incomplete (heuristic) proofs—as opposed to a critique of the FEP per se The issues they identify are still fundamental. The novel contribution of this paper is an explicit specification of the conditions imposed upon a dynamical flow that are sufficient to ensure a Markov blanket
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
