nucleation

THEORY-1245: Evaluating expected free energy recursively over the beliefs an agent would hold after each plausible outcome makes active inference a pruned tree search over belief states, formally analogous to Bellman recursion, which lets the agent plan through short-term costs that one-step agents cannot; that the pruning keeps deep search tractable beyond toy problems is argued, not shown

1 min read

Properties4
StatusProposed
Tagsprobabilistic-modeling, information-theory
SourceLIT-577 — Sophisticated Inference
ExtendsTHEORY-426 — Evaluating expected free energy recursively over the beliefs an agent would hold after each plausible outcome makes active inference a pruned tree search over belief states, formally analogous to Bellman recursion; in single deterministic runs on a T-maze and an 8 × 8 maze it planned through short-term costs that agents with a shorter horizon did not; that the pruning keeps deep search tractable beyond toy problems is argued, not shown

Graph View

Source

Friston et al. (2020), “Sophisticated Inference”, LIT-577, read in full as prose in NOTE-454. What the paper’s runs showed is THEORY-426, which this entry extends.

Why it is Proposed

The title states the planning claim generally, where the paper shows it on toy problems, “rather trivial problems” by the authors’ own account. The cost figure, “usually several hundred” of about 10¹⁰ or 10¹⁵ paths, is for one maze. A 1/16 pruning threshold cuts little where many outcomes stay plausible, and no such problem is tried (NOTE-454, Open questions). The paper says the scheme recovers Bayes-adaptive RL at zero temperature, but no Bayes-adaptive or Monte Carlo planner is run on the same tasks.

Backlinks