Conformal Sequential Decision Making With Cost and Sublinear Policy Regret Guarantees

The Online Conformal Episodic Agent (OCEA) is a policy designed to deterministically meet per-episode cost constraints while achieving sublinear expected policy regret. It works by selecting one candidate action per episode and switching to a zero-cost action if the remaining budget cannot cover the maximum potential cost. This ensures safety in sequential decision-making problems.

Open the full episode

More from AI Radio