Reinforcement Learning with Conformal Action Sets: An Application to Sequential Recommendation
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…
We haven't written up this one. arXiv cs.AI has the full story — the link below goes straight to it.
