gin-factorial-hmm-hidden-state-structure

IN premise — summaries/2026/08/24/xie-2021-icl-bayesian-sR-references-chunk-1.md

Created 2026-08-25T02:58:56+00:00

GINC uses a factorial HMM structure with 10 entities × 10 properties = 100 hidden states, 5 latent concepts, entity self-transition probability ≥ 0.9, and deterministic emissions via a memory matrix M[v,s] (p(o_t|h_t) = 1 if o_t = M[v_t, s_t]).

Summary

The GINC model is built to be very stable and predictable: once an entity settles into a particular property, it almost always stays there, and whatever it outputs at any step is a fixed lookup rather than a random draw. This means the system trades flexibility for determinism, producing consistent and reproducible outputs, but any small error in the underlying state assignments gets locked in rather than smoothed out by probability.