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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add dual-critic PAMDP RL on top of behavior cloning | BC actor · PrefEval (Qwen2.5-7B) | r_w: 0.296 -> 0.439 | Improved |
| add dual-critic PAMDP RL on top of behavior cloning | BC actor · PrefEval (Llama3-8B) | r_w: 0.3265 -> 0.3776 | Improved |
| add dual-critic PAMDP RL on top of behavior cloning | BC actor · ALOE (Qwen2.5-7B) | r_w: 0.0809 -> 0.1046 | Improved |
| add dual-critic PAMDP RL on top of behavior cloning | BC actor · ALOE (Llama3-8B) | r_w: 0.1945 -> 0.2671 | Improved |
| add chain-of-thought profile inference to full fine-tuning | FPFT · PrefEval (Qwen2.5-7B) | r_w: 0.1020 -> 0.1326 | Improved |
| add RL (actor-critic) on top of BC (ablation) | BC · DeepSeek-R1 judge, vs Vanilla | r_w: 0.4050 -> 0.4300 | Improved |
| add profile as unobservable state variable in POMDP advantage (ablation) | RL-base · DeepSeek-R1 judge, vs Vanilla | r_w: 0.4300 -> 0.4655 | Improved |
| use dual-critic instead of asymmetric UAAC | UAAC · Online ALOE, step 6 return | average return: 1.5784 -> 1.7389 | Improved |
| use dual-critic instead of DCRL | DCRL · Online ALOE, step 6 return | average return: 1.4305 -> 1.7389 | Improved |
| use dual-critic PAMDP instead of GRPO | GRPO · judge Qwen3-30B-A3B | r_w: 0.4097 -> 0.4352 | Improved |
| use dual-critic PAMDP instead of DPO | DPO · judge DeepSeek-R1-0528-AWQ | r_w: 0.4546 -> 0.4655 | Improved |
| use user-profile-aware prompting only | Vanilla LLM · ALOE (Qwen2.5-7B) | r_w: 0.0467 | No clear effect |
Closest papers by meaning
- Consistently Simulating Human Personas with Multi-Turn Reinforcement Learning (NeurIPS 2025 · similarity 0.91)
- Teaching Language Models to Evolve with Users: Dynamic Profile Modeling for Personalized Alignment (NeurIPS 2025 · similarity 0.89)
- Enhancing Personalized Multi-Turn Dialogue with Curiosity Reward (NeurIPS 2025 · similarity 0.87)
- PersonaVLM: Long-Term Personalized Multimodal LLMs (CVPR 2026 · similarity 0.86)
- Who's asking? User personas and the mechanics of latent misalignment (NeurIPS 2024 · similarity 0.85)
- Persona Vectors: Monitoring and Controlling Character Traits in Language Models (NeurIPS 2026 · similarity 0.85)
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.
