Name: Active Inference
Author: Solaris-CET

SKILL: Active Inference

Overview

A Partially Observable Markov Decision Process (POMDP) engine that selects actions by minimising Variational Free Energy rather than maximising expected reward.

When to apply

Designing or reviewing agents that maintain beliefs over hidden state.
Choosing between explore (reduce uncertainty) vs exploit (satisfy preferences).
Implementing or validating A, B, C, D matrix contracts and normalisation.

Agent workflow

Represent the problem with explicit states, observations, and actions where possible.
After each observation, update beliefs with Bayes-style fusion (log-space softmax pattern below).
Score candidate actions with ; prefer lower EFE unless the user specifies otherwise.

SKILL: Active Inference

Overview

A Partially Observable Markov Decision Process (POMDP) engine that selects actions by minimising Variational Free Energy rather than maximising expected reward.

When to apply

Designing or reviewing agents that maintain beliefs over hidden state.
Choosing between explore (reduce uncertainty) vs exploit (satisfy preferences).
Implementing or validating A, B, C, D matrix contracts and normalisation.

Agent workflow

Represent the problem with explicit states, observations, and actions where possible.
After each observation, update beliefs with Bayes-style fusion (log-space softmax pattern below).
Score candidate actions with ; prefer lower EFE unless the user specifies otherwise.

Matrix	Shape	Semantics
A	`[O × S]`	Likelihood: `P(observation
B	`[S × S × A]`	Transition: `P(next_state
C	`[O]`	Log-preferences over observations (goal encoding)
D	`[S]`	Prior belief over initial states

Active Inference

SKILL: Active Inference

Overview

When to apply

Agent workflow

Active Inference

SKILL: Active Inference

Overview

When to apply

Agent workflow

POMDP Matrices

Free Energy Minimisation

Axiom

Matematico Tao

Seo Fundamentals

Yann Lecun Debate

Yann Lecun Filosofia

Explain Like Socrates