Empirical Validation with Learned or Implicit Belief States

Validate the Triadic Cognitive Architecture empirically in agents whose belief states are learned or implicit and must be approximated from model outputs or sampled posteriors.

Background

The main experiments use categorical beliefs with Dirichlet-likelihood updates that admit closed-form Bayesian inference. In many practical systems, beliefs are implicit or learned and require approximation from model outputs or sampled posteriors.

While the authors outline one possible bridge for such settings, they explicitly state that empirical validation on agents with implicit or learned belief states remains future work.