The Simple Rule Won Under Random Hazards

A two-action causal rule based on local nonlinearity decisively beat the corrected neural controller on fresh stochastic filtering rollouts.

ai-research, bayesian-filtering, nonlinear-filtering, robustness, heuristics