T-REG: Preference Optimization with Token-Level Reward Regularization (2025.acl-long)
Copied to clipboard
| Challenge: | Reinforcement learning from human feedback (RLHF) is a dominant approach for large language models to follow instructions and produce meaningful alignment. |
| Approach: | They propose a method that leverages human feedback to optimize large language models . they propose to use sequence-level and token-level rewards to optimize preference . |
| Outcome: | The proposed method outperforms baseline methods on Alpaca Eval 2 and Arena-Hard benchmarks. |
Similar Papers
TLCR: Token-Level Continuous Reward for Fine-grained Reinforcement Learning from Human Feedback (2024.findings-acl)
Copied to clipboard
Eunseop Yoon, Hee Suk Yoon, SooHwan Eom, Gunsoo Han, Daniel Nam, Daejin Jo, Kyoung-Woon On, Mark Hasegawa-Johnson, Sungwoong Kim, Chang Yoo
| Challenge: | Existing approaches to provide token-level rewards fail to account for varying degrees of preference inherent to each token. |
| Approach: | They propose a reward model that uses a discriminator to assign token-based continuous rewards to each token considering the context. |
| Outcome: | Extensive experiments show that the proposed reward model improves on open-ended language generation benchmarks. |
RED: Unleashing Token-Level Rewards from Holistic Feedback via Reward Redistribution (2025.emnlp-main)
Copied to clipboard
| Challenge: | Experimental results demonstrate the superiority of our approach to aligning large language models with human preferences. |
| Approach: | They propose a method that evaluates and assigns specific credit to each token using an off-the-shelf reward model. |
| Outcome: | The proposed method evaluates and assigns specific credit to each token using an off-the-shelf reward model. |
AlignDistil: Token-Level Language Model Alignment as Adaptive Policy Distillation (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for LLM alignment optimize tokens using a sparse, response-level reward or preference annotation. |
| Approach: | They propose an RLHF-equivalent distillation method for token-level reward optimization that incorporates the reward learned by DPO into the RLHG objective and builds a token-based teacher distribution. |
| Outcome: | The proposed method bridges the accuracy gap between the reward from the DPO model and the pure reward model by building a contrastive DPO reward with a normal and a reverse DPO. |
LIRE: listwise reward enhancement for preference alignment (2024.findings-acl)
Copied to clipboard
| Challenge: | prevailing approaches to preference alignment focus on pairwise comparisons, with limited exploration into multi-response scenarios. |
| Approach: | They propose a listwise reward enhancement approach that integrates offline rewards of multiple responses into a streamlined listwise framework. |
| Outcome: | The proposed approach outperforms existing methods on dialogue and summarization tasks with good transferability to out-of-distribution data. |
Prototypical Reward Network for Data-Efficient Model Alignment (2024.acl-long)
Copied to clipboard
| Challenge: | Reinforcement Learning from Human Feedback (RLHF) is a reward model that fine-tunes Large Language Models (LLMs) by utilizing Prototypical Networks. |
| Approach: | They propose a framework utilizing Prototypical Networks to enhance reward models under limited human feedback, enabling more stable and reliable structural learning from fewer samples. |
| Outcome: | The proposed framework improves reward models under limited human feedback, surpassing traditional methods, especially in data-limited scenarios. |
Selective Preference Optimization via Token-Level Reward Function Estimation (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing methods for maximizing preference optimization on all available tokens are noisy and inefficient. |
| Approach: | They propose a selective alignment strategy that centers on efficient key token selection without strong, fine-grained supervision signals. |
| Outcome: | The proposed strategy outperforms baseline methods on three benchmarks with up to 60% reduction in training hours. |
Inverse-Q*: Token Level Reinforcement Learning for Aligning Large Language Models Without Preference Data (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Reinforcement Learning from Human Feedback (RLHF) relies on complex methodologies like Proximal Policy Optimization (PPO) that require extensive hyper-parameter tuning and pose challenges in sample efficiency and stability. |
| Approach: | They propose an innovative framework that leverages direct preference optimization techniques but extends them by estimating the conditionally optimal policy directly from the model’s responses. |
| Outcome: | The proposed framework matches and exceeds the effectiveness of Proximal Policy Optimization (PPO) in terms of convergence speed and alignment of model responses with human preferences. |
ARF-RLHF: Adaptive Reward-Following for RLHF through Emotion-Driven Self-Supervision and Trace-Biased Dynamic Optimization (2026.acl-long)
Copied to clipboard
| Challenge: | prevailing RLHF methods such as PPO and DPO depend on large-scale binary preference annotations. |
| Approach: | They propose a method which converts natural feedback into continuous preference trajectories and optimizes them using the novel TraceBias algorithm. |
| Outcome: | The proposed approach outperforms PPO and DPO in a variety of domains and improves alignment by up to 7.6% across diverse LLMs and preference domains. |
Aligning Large Language Models via Fine-grained Supervision (2024.acl-short)
Copied to clipboard
| Challenge: | Pre-trained large-scale language models often generate biased or toxic text, misaligning with human intentions. |
| Approach: | They propose to use human feedback to improve LLM alignment by fine-grained token supervision . they ask annotators to edit less preferred responses to make them more favorable . |
| Outcome: | The proposed method improves LLM alignment by up to 5.1% in terms of win rate compared with the traditional model. |
Self-supervised Preference Optimization: Enhance Your Language Model with Preference Degree Awareness (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Recent studies have focused on replacing the reward model in Reinforcement Learning with Human Feedback (RLHF) methods for Large Language Models (LLMs). |
| Approach: | They propose a self-supervised preference optimization framework that replaces the reward model with a preference loss and alignment loss to improve LLMs' ability to understand human preferences. |
| Outcome: | The proposed framework can be integrated with existing preference optimization methods and significantly boost their performance. |