Reinforcement Learning with Conformal Action Sets: An Application to Sequential Recommendation

Excerpt

arXiv:2610.08743v1 Announce Type: new Abstract: Sequential recommenders typically use a fixed slate size even though the number of useful alternatives changes within a session. We propose Reinforcement Learning with Calibrated Pruning (RLCP), which adapts the retained action set using critic scores and an online threshold. The threshold is updated from binary feedback indicating whether the set contains an action in a proxy target. We prove a deterministic bound on the observed proxy miss rate a