Research map / Large language models / Language, safety and interpretability
From Tokens to Latent States: Leveraging Pre-trained Language Models for Improving Partially Observable Reinforcement Learning
Meiju Li, Ruixiang Sun, Xin Li, Mingzhong Wang
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 |
|---|---|---|---|
| use frozen pre-trained TransformerXL belief module (ELSLLM) instead of PPO without memory | PPO · Minigrid Unlock | MSE of estimated state-values: 0.0207 -> 0.0004 | Improved |
| use ELSLLM instead of LSTM-PPO memory | LSTM-PPO · Minigrid Unlock | MSE of estimated state-values: 0.1686 -> 0.0004 | Improved |
| use ELSLLM (actions plus observations, JLP) instead of HELM (visual observations only) | HELM · Minigrid Unlock | MSE of estimated state-values: 0.0098 -> 0.0004 | Improved |
| add JLP and MHN input mapping instead of learnable MLP input layers (ablation) | ELSLLM · Minigrid Unlock | episode reward | Improved |
| add pre-trained TransformerXL weights instead of random initialization (ablation) | ELSLLM · Minigrid Unlock | episode reward | Improved |
| add LSTM memory to PPO | PPO · Minigrid (7 tasks) | episode reward | Mixed |
| add LSTM memory to PPO | PPO · RandomMaze | episode reward | Hurt |
Closest papers by meaning
- Efficient Reinforcement Learning with Large Language Model Priors (ICLR 2025 · similarity 0.90)
- Learning POMDP World Models from Observations with Language-Model Priors (NeurIPS 2026 · similarity 0.89)
- Controlling Large Language Model with Latent Action (ICML 2025 · similarity 0.89)
- Mitigating Partial Observability in Sequential Decision Processes via the Lambda Discrepancy (NeurIPS 2024 · similarity 0.89)
- On Predictability of Reinforcement Learning Dynamics for Large Language Models (ICLR 2026 · similarity 0.89)
- Reinforcement World Model Learning for LLM-based Agents (NeurIPS 2026 · similarity 0.89)
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.
