AI Paper Atlas

Research map / Reinforcement learning / MDPs and planning

PAMDP: Interact to Persona Alignment via a Partially Observable Markov Decision Process

ZHE YANG, Yi Huang, Si Chen, Xiaoting Wu, Jingyu Yao, Junlan Feng

ICLR 2026 · Reinforcement learning › MDPs and planning · approach: Average-reward and regularized MDPs

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
add dual-critic PAMDP RL on top of behavior cloningBC actor · PrefEval (Qwen2.5-7B)r_w: 0.296 -> 0.439Improved
add dual-critic PAMDP RL on top of behavior cloningBC actor · PrefEval (Llama3-8B)r_w: 0.3265 -> 0.3776Improved
add dual-critic PAMDP RL on top of behavior cloningBC actor · ALOE (Qwen2.5-7B)r_w: 0.0809 -> 0.1046Improved
add dual-critic PAMDP RL on top of behavior cloningBC actor · ALOE (Llama3-8B)r_w: 0.1945 -> 0.2671Improved
add chain-of-thought profile inference to full fine-tuningFPFT · PrefEval (Qwen2.5-7B)r_w: 0.1020 -> 0.1326Improved
add RL (actor-critic) on top of BC (ablation)BC · DeepSeek-R1 judge, vs Vanillar_w: 0.4050 -> 0.4300Improved
add profile as unobservable state variable in POMDP advantage (ablation)RL-base · DeepSeek-R1 judge, vs Vanillar_w: 0.4300 -> 0.4655Improved
use dual-critic instead of asymmetric UAACUAAC · Online ALOE, step 6 returnaverage return: 1.5784 -> 1.7389Improved
use dual-critic instead of DCRLDCRL · Online ALOE, step 6 returnaverage return: 1.4305 -> 1.7389Improved
use dual-critic PAMDP instead of GRPOGRPO · judge Qwen3-30B-A3Br_w: 0.4097 -> 0.4352Improved
use dual-critic PAMDP instead of DPODPO · judge DeepSeek-R1-0528-AWQr_w: 0.4546 -> 0.4655Improved
use user-profile-aware prompting onlyVanilla LLM · ALOE (Qwen2.5-7B)r_w: 0.0467No clear effect

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