Provably Safe Offline-to-Online RL: Decoupling Learning from Data-Driven Safety Enforcement (2026.acl-long)
Copied to clipboard
| Challenge: | Hybrid offline–online reinforcement learning (O2O RL) promises both sample efficiency and robust exploration, but suffers from instability due to distribution shift between offline and online data. |
| Approach: | They propose a framework that decouples policy optimization from safety enforcement . they propose dynamic curricula that gradually extend temporal horizons and anneal offline–online data mixing . |
| Outcome: | The proposed framework preserves the exploratory value of online interactions without collapsing to conservative policies. |
Similar Papers
SOLAR-RL: Semi-Online Long-horizon Assignment Reinforcement Learning (2026.findings-acl)
Copied to clipboard
Jichao Wang, Liuyang Bian, Yufeng Zhou, Han Xiao, Yue Pan, Guozhi Wang, Hao Wang, Zhaoxiong Wang, Yafei Wen, Xiaoxin Chen, Shuai Ren, Lingfang Zeng
| Challenge: | Existing approaches to training GUI agents on dynamic tasks are based on SFT or Behavior Cloning. |
| Approach: | They propose a framework that integrates global trajectory insights directly into offline learning . they reconstruct diverse rollout candidates from static data and detect first failure point . |
| Outcome: | The proposed framework improves long-horizon task completion rates and robustness compared to baselines. |
MORE-3S:Multimodal-based Offline Reinforcement Learning with Shared Semantic Spaces (2024.lrec-main)
Copied to clipboard
| Challenge: | Existing approaches to offline reinforcement learning (RL) focus on learning value functions or policy gradients, but they view it as a sequence modeling task. |
| Approach: | They propose a method that integrates multimodal and pre-trained language models to transform offline reinforcement learning into a supervised learning task by integrating state information derived from images and action-related data obtained from text. |
| Outcome: | The proposed approach outperforms baselines on Atari and OpenAI Gym environments while promoting long-term strategic thinking. |
TriPlay-RL: Tri-Role Self-Play Reinforcement Learning for LLM Safety Alignment (2026.acl-long)
Copied to clipboard
Zhewen Tan, Wenhan Yu, Jianfeng Si, Tongxin Liu, Kaiqi Guan, Huiyan Jin, Jiawen Tao, Xiaokun Yuan, Xiangzheng Zhang, Duohe Ma, Tong Yang, Lin Sun
| Challenge: | Existing approaches to safety alignment of large language models rely on costly manual annotations or human review. |
| Approach: | They propose a closed-loop reinforcement learning framework called TriPlay-RL that enables iterative collaboration among three roles with near-zero manual annotation. |
| Outcome: | The proposed framework achieves 20%–50% improvement in adversarial effectiveness while preserving high output diversity while achieving 10%–30% gains in safety performance without degrading general reasoning capability. |
Human-centric dialog training via offline reinforcement learning (2020.emnlp-main)
Copied to clipboard
Natasha Jaques, Judy Hanwen Shen, Asma Ghandeharioun, Craig Ferguson, Agata Lapedriza, Noah Jones, Shixiang Gu, Rosalind Picard
| Challenge: | a novel offline RL method can train dialog models to produce better conversations without the risk of humans teaching it harmful chat behaviors. |
| Approach: | They develop offline reinforcement learning algorithms that use human feedback to train dialog models . they use language similarity, laughter, sentiment, and more to identify positive feedback . |
| Outcome: | The proposed method improves on existing methods with 80 users in an open-domain setting. |
Scalable and Safe Remediation of Defective Actions in Self-Learning Conversational Systems (2023.acl-industry)
Copied to clipboard
| Challenge: | Off-Policy reinforcement learning has been used to improve conversational AIs, but in large-scale commercial environments it is challenging to balance between policy improvements and experience continuity. |
| Approach: | They propose to curate and leverage regression incident reports to validate, safe-guard, and improve policies prior to the online deployment. |
| Outcome: | The proposed method validates, safe-guards, and improves policies prior to the online deployment. |
trlX: A Framework for Large Scale Reinforcement Learning from Human Feedback (2023.emnlp-main)
Copied to clipboard
Alexander Havrilla, Maksym Zhuravinskyi, Duy Phung, Aman Tiwari, Jonathan Tow, Stella Biderman, Quentin Anthony, Louis Castricato
| Challenge: | Current RLHF paradigms rely on Proximal Policy Optimization (PPO), which quickly becomes a challenge to implement and scale up to large architectures. |
| Approach: | They propose an open-source framework for reinforcement learning from human feedback . it allows for offline fine-tuning of large language models . |
| Outcome: | The framework can be used to fine-tune models up to and exceeding 70 billion parameters. |
Mobile-R1: Towards Interactive Capability for VLM-Based Mobile Agent via Systematic Training (2026.acl-long)
Copied to clipboard
Jihao Gu, Qihang Ai, Yingyao Wang, Pi Bu, Jingxuan Xing, Yue Cao, Zekun Zhu, Wei Jiang, Ziming Wang, Yingxiu Zhao, Ming-Liang Zhang, Jun Song, Yuning Jiang, Bo Zheng
| Challenge: | Existing approaches to training agents for visual-language models trap them in local optima, hindering exploration and error correction with the environment. |
| Approach: | They propose a hierarchical training recipe that bridges atomic action execution and strategic task completion. |
| Outcome: | The proposed training recipe bridges atomic action execution and strategic task completion. |
From log 𝜋 to 𝜋: Taming Divergence in Soft Clipping via Bilateral Decoupled Decay of Probability Gradient Weight (2026.acl-long)
Copied to clipboard
Xiaoliang Fu, Jiaye Lin, Yangyi Fang, Chaowen Hu, Cong Qin, Zekai Shao, Binbin Zheng, Lu Pan, Ke Zeng
| Challenge: | Standard algorithms for Large Language Models (LLMs) enforce stability via "hard clipping" but relying on log-probability gradient yields divergent weights as probabilities vanish, destabilizing LLM training. |
| Approach: | They propose a decoupled gradient policy optimization that uses a decay mechanism to decouple the probability of a boundary token. |
| Outcome: | The proposed algorithm outperforms baselines on various mathematical benchmarks. |
Targeted Exploration via Unified Entropy Control for Reinforcement Learning (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for group relative policy optimization suffer from entropy collapse . Existing exploration methods introduce additional bias or variance during exploration, making it difficult to maintain stability. |
| Approach: | They propose a framework that provides targeted mechanisms for exploration and stabilization. |
| Outcome: | The proposed framework expands search space on difficult prompts while preventing entropy growth uncontrollably. |
Reflect-RL: Two-Player Online RL Fine-Tuning for LMs (2024.acl-long)
Copied to clipboard
| Challenge: | supervised fine-tuning (SFT) on a limited offline dataset does not yield good performance. |
| Approach: | They propose a two-player system to fine-tune an LM using SFT and online RL . they use negative example generation to enhance error-correction ability of the reflection model . |
| Outcome: | The proposed system outperforms SFT and online RL without reflection on a GPT-2 XL 1.56B model. |