AI Paper Atlas

Research map / Reinforcement learning / Offline RL

Offline Reinforcement Learning for Mixture-of-Expert Dialogue Management

Dhawal Gupta, Yinlam Chow, Azamat Tulepbergenov, Mohammad Ghavamzadeh, Craig Boutilier

NeurIPS 2023 · Reinforcement learning › Offline RL · approach: Offline-to-online fine-tuning · 1 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Use model-based evaluation with a learned next-user model rather than model-free Q scoringoffline RL dialogue managers · Reddit Casual and CornellreturnImproved
Use MoE-specific SAIQL with all expert candidate utterances rather than the best standard offline RL methodMoE-LM dialogue management · Reddit Casual, model-basedreturn: 4.25 ± 0.12 -> 4.65 ± 0.06Improved
Use MoE-specific FtLE rather than the best standard offline RL methodMoE-LM dialogue management · Cornell, model-freereturn: -1.32 ± 0.19 -> -0.39 ± 0.24Improved
Use MoE-specific MoE-VRL rather than the best standard offline RL methodMoE-LM dialogue management · Cornell, model-basedreturn: 1.47 ± 0.15 -> 3.62 ± 0.17Improved
Use MoE-specific methods that incorporate all experts rather than the best standard offline RL methodsMoE-LM dialogue management · Reddit Casual and CornellreturnMixed
Use FtLE or MoE-VRL, which model expert-specific long-term value, instead of SAIQLMoE-LM dialogue management · Cornell, model-free and model-basedreturnImproved
Use IQL rather than SAC, EnsQ, or KLC among standard offline RL methodsMoE primitive LM dialogue management · most model-based and model-free evaluationsreturnMixed
Use IQL rather than BCoffline dialogue management · Reddit Casual and CornellreturnImproved
Use standard offline RL rather than per-step reward-maximizing Banditsstandard offline RL dialogue managers · Reddit Casual and CornellreturnNo clear effect
Use MoE-aware offline RL rather than per-step Bandit greedificationMoE-LM dialogue management · Reddit Casual and CornellreturnImproved
Use SAC rather than EnsQ or KLCstandard offline RL dialogue management · Reddit Casual, model-freereturnMixed
Apply dropout-based ensemble Q or KL regularization in the fixed MoE latent spaceEnsQ and KLC · open-domain dialogue managementreturnHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy