xin-icl-deterministic-emissions

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

Created 2026-08-25T02:58:57+00:00

GINC uses a factorial HMM with deterministic emissions where p(o_t | h_t) = 1 iff o_t = M[v_t, s_t], isolating the property transition matrix as the only varying parameter.

Summary

GINC's output at each step is a fixed lookup determined entirely by its hidden state — there is no randomness or noise in what it emits, only one specific value is possible for any given state. This means the only thing the model can actually learn or adapt over time is how it moves between states, making the learning problem far simpler and easier to reason about than a model where both transitions and emissions are uncertain.