Papers by Xi Niu
Adversarial Preference Learning for Robust LLM Alignment (2025.findings-acl)
Copied to clipboard
Yuanfu Wang, Pengyu Wang, Chenyang Xi, Bo Tang, Junyi Zhu, Wenqiang Wei, Chen Chen, Chao Yang, Jingfeng Zhang, Chaochao Lu, Yijun Niu, Keming Mao, Zhiyu Li, Feiyu Xiong, Jie Hu, Mingchuan Yang
| Challenge: | Modern language models rely on Reinforcement Learning from Human Feedback (RLHF) to encourage safe behaviors, but they remain vulnerable to adversarial attacks due to three key limitations: (1) the inefficiency and high cost of human annotation; (2) the vast diversity of potential adversarials; and (3) the risk of feedback bias and reward hacking. |
| Approach: | They propose an iterative adversarial training method that incorporates three key innovations to address these challenges. |
| Outcome: | Experiments on Mistral-7B-Instruct-v0.3 show that the proposed method significantly enhances robustness and reduces harmful outputs from 5.88% to 0.43%. |
Fine-tuning LLMs with Cross-Attention-based Weight Decay for Bias Mitigation (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) excel in natural language processing tasks but often propagate societal biases from their training data, leading to discriminatory outputs. |
| Approach: | They propose a method that modifies the LLM architecture to mitigate bias by adjusting the attention weights of sensitive tokens. |
| Outcome: | The proposed method can handle multiple sensitive attributes and does not require full knowledge of sensitive tokens presented in the dataset. |
Sheaf Discovery with Joint Computation Graph Pruning and Flexible Granularity (2025.emnlp-main)
Copied to clipboard
| Challenge: | Experimental results show that DiscoGP extracts sheaves that preserve 93-100% of a model’s performance while comprising only 1-7% of the original weights and connections. |
| Approach: | They propose a framework for extracting self-contained modular units within neural language models (LMs) they use a gradient-based pruning algorithm to prune the original LM to a sparse skeleton . |
| Outcome: | The proposed framework preserves 93-100% of the original model's performance while preserving only 1-7% of the model''s original weights and connections. |
Fair RAG: End-to-End Fairness Across Retrieval and Generation (2026.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) can amplify demographic bias by generating skewed context . prior work treats fairness in retrieval or generation in isolation, leaving end-to-end fairness underexplored . |
| Approach: | They propose a pipeline that jointly controls both retrieval and generation stages . large language models can handle a broad set of inference tasks, they argue . |
| Outcome: | The proposed pipeline reduces retriever-side skew and achieves lowest generator-side disparity while preserving utility. |