Papers by Shujun Wang
HAF-RM: A Hybrid Alignment Framework for Reward Model Training (2025.acl-long)
Copied to clipboard
Shujun Liu, Xiaoyu Shen, Yuhang Lai, Siyuan Wang, Shengbin Yue, Zengfeng Huang, Xuanjing Huang, Zhongyu Wei
| Challenge: | Recent studies have focused on enhancing reward models through data improvements, following the conventional training framework for reward models that directly optimizes the predicted rewards. |
| Approach: | They propose a hybrid alignment framework **HAF-RM** that incorporates additional constraint on token-level policy probabilities in addition to the reward score. |
| Outcome: | The proposed framework can supervise the internal preference model at the token level and optimize the mapping layer of the reward model at sequence level. |
ALaRM: Align Language Models via Hierarchical Rewards Modeling (2024.findings-acl)
Copied to clipboard
| Challenge: | Current alignment approaches struggle with inconsistency and sparsity of human supervision signals. |
| Approach: | They propose a framework modeling hierarchical rewards in reinforcement learning from human feedback (RLHF) it integrates holistic rewards with aspect-specific rewards to enhance alignment of large language models with human preferences. |
| Outcome: | The proposed framework improves the alignment of large language models with human preferences by integrating holistic rewards with aspect-specific rewards. |
Multi-Agent Simulator Drives Language Models for Legal Intensive Interaction (2025.findings-naacl)
Copied to clipboard
ShengbinYue ShengbinYue, Ting Huang, Zheng Jia, Siyuan Wang, Shujun Liu, Yun Song, Xuanjing Huang, Zhongyu Wei
| Challenge: | Large Language Models (LLMs) have advanced legal intelligence, but the scarcity of scenario data impedes the progress toward interactive legal scenarios. |
| Approach: | They propose a Multi-agent Legal Simulation Driver to generate synthetic data by simulating interactive legal scenarios. |
| Outcome: | The proposed framework ensures consistency of legal attributes between participants and introduces a supervisory mechanism to align participants’ characters and behaviors as well as addressing distractions. |
Guided by the Plan: Enhancing Faithful Autoregressive Text-to-Audio Generation with Guided Decoding (2026.eacl-long)
Copied to clipboard
| Challenge: | Autoregressive (AR) models excel at generating temporally coherent audio by producing tokens sequentially, yet they often falter in faithfully following complex textual prompts. |
| Approach: | They propose a lightweight auxiliary model trained with a GAE-inspired objective to predict final instruction-following quality from partial generations. |
| Outcome: | The proposed model achieves 10 points improvement in CLAP score over baseline AR models while maintaining computational parity with best-of-N decoding. |
Language Model Based Text-to-Audio Generation: Anti-Causally Aligned Collaborative Residual Transformers (2025.emnlp-main)
Copied to clipboard
| Challenge: | Autoregressive language models excel in text-to-audio generation, but lag behind diffusion models by a non-trivial margin. |
| Approach: | They propose a framework that integrates multiple isolated transformers with causal conditioning and anti-causal alignment via reinforcement learning. |
| Outcome: | The proposed framework outperforms existing LM-based and diffusion-based systems in audio synthesis. |
How Jailbreak Defenses Work and Ensemble? A Mechanistic Investigation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Jailbreak attacks, where harmful prompts bypass generative models’ built-in safety, raise serious concerns about model vulnerability. |
| Approach: | They propose to reframe the standard generation task as a binary classification problem to assess model refusal tendencies for both harmful and benign queries. |
| Outcome: | The proposed defenses improve model safety or optimize the trade-off between safety and helpfulness. |