Papers by Haiyan Zhao
The Elephant in the Room: Exploring the Role of Neutral Words in Language Model Group-Agnostic Debiasing (2025.findings-acl)
Copied to clipboard
Xinwei Guo, Jiashi Gao, Junlei Zhou, Jiaxin Zhang, Guanhua Chen, Xiangyu Zhao, Quanying Liu, Haiyan Wu, Xin Yao, Xuetao Wei
| Challenge: | Large Language Models (LLMs) are increasingly integrated into our daily lives, raising ethical concerns, especially about perpetuating stereotypes. |
| Approach: | They propose a method that incorporates a neutral word semantics-based loss function to alleviate the deterioration of the LMS during debiasing. |
| Outcome: | The proposed method alleviates the deterioration of the Language Modeling Score (LMS) by incorporating a neutral word semantics-based loss function. |
The Impact of Reasoning Step Length on Large Language Models (2024.findings-acl)
Copied to clipboard
| Challenge: | Long reasoning steps in LLMs improve reasoning abilities, but the correlation between their effectiveness and the length of reasoning steps remains largely unknown. |
| Approach: | They conducted experiments that expand and compress the rationale reasoning steps within CoT demonstrations while keeping all other factors constant. |
| Outcome: | The results show that lengthening the reasoning steps in prompts significantly enhances LLMs’ reasoning abilities across multiple datasets. |
LLMs Trust Humans More, That’s a Problem! Unveiling and Mitigating the Authority Bias in Retrieval-Augmented Generation (2025.acl-long)
Copied to clipboard
Yuxuan Li, Xinwei Guo, Jiashi Gao, Guanhua Chen, Xiangyu Zhao, Jiaxin Zhang, Quanying Liu, Haiyan Wu, Xin Yao, Xuetao Wei
| Challenge: | Large language models (LLMs) generate outputs that stray from user input or contravene established knowledge. |
| Approach: | They propose a new phenomenon, Authority Bias, where LLMs favor one knowledge source over the other . they propose atomic information that generates conflicts and a Conflict Detection Enhanced Query framework . |
| Outcome: | The proposed framework reduces Authority bias in large language models . it detects conflicts, performs credibility assessment on conflicting paragraphs, and detects perturbed text . |
Mitigating Shortcuts in Language Models with Soft Label Encoding (2024.lrec-main)
Copied to clipboard
| Challenge: | Recent studies have shown that large language models rely on spurious correlations in the data for natural language understanding (NLU) tasks. |
| Approach: | They propose a framework for debiasing shortcuts and a dummy class to encode shortcuts into a model and use it to generate soft labels. |
| Outcome: | The proposed framework significantly improves out-of-distribution generalization while maintaining satisfactory in-district accuracy. |
A Survey on Sparse Autoencoders: Interpreting the Internal Mechanisms of Large Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Sparse Autoencoders (SAEs) can disentangle complex features into more interpretable components. |
| Approach: | They propose to use Sparse Autoencoders to disentangle LLM features into more interpretable components. |
| Outcome: | The proposed method disentangles complex features into more interpretable components. |
EVIT: Event-Oriented Instruction Tuning for Event Reasoning (2024.findings-acl)
Copied to clipboard
| Challenge: | Large language models (LLMs) have made significant advances in event reasoning . however, smaller instruction-tuned models do not consistently demonstrate exceptional proficiency . |
| Approach: | They propose an event-oriented instruction tuning technique to train a large language model . they propose a structure named event quadruple which contains the structure and semantics of events . |
| Outcome: | The proposed model achieves competitive performances on event reasoning tasks. |
Can LLMs Hear the Dogwhistle? (2026.findings-acl)
Copied to clipboard
Yifan Liu, Yi Lin, Xinwei Guo, Ziwei Wang, Jiaxin Zhang, Guanhua Chen, Haiyan Wu, Xiangyu Zhao, Xin Yao, Xuetao Wei
| Challenge: | Existing safety benchmarks focus on explicitly harmful content, but ignore context-dependent expressions such as dogwhistles. |
| Approach: | They propose a benchmark for evaluating LLM safety under dogwhistle-driven prompts . their findings expose a blind spot in current safety evaluation practices . |
| Outcome: | The proposed benchmark compared safety performance with toxic terms using dogwhistle-driven prompts. |
Large Vision-Language Model Alignment and Misalignment: A Survey Through the Lens of Explainability (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Large Vision-Language Models have demonstrated remarkable capabilities in processing both visual and textual information. |
| Approach: | They examine the challenge of alignment and misalignment in LVLMs through an explainability lens. |
| Outcome: | The findings highlight the need for standardized evaluation protocols and in-depth explainability studies. |
Exploring Concept Depth: How Large Language Models Acquire Knowledge and Concept at Different Layers? (2025.coling-main)
Copied to clipboard
Mingyu Jin, Qinkai Yu, Jingyuan Huang, Qingcheng Zeng, Zhenting Wang, Wenyue Hua, Haiyan Zhao, Kai Mei, Yanda Meng, Kaize Ding, Fan Yang, Mengnan Du, Yongfeng Zhang
| Challenge: | Large language models have shown remarkable performances across a wide range of tasks, but mechanisms by which they encode tasks of varying complexity remain poorly understood. |
| Approach: | They propose to explore the possibility that LLMs process concepts in different layers . they propose to categorize concepts based on their level of abstraction . |
| Outcome: | The proposed model can process complex concepts in shallow layers, the authors show . the proposed model could be used to prob complex tasks in shallow ones . |
Detection, Diagnosis, and Explanation: A Benchmark for Chinese Medial Hallucination Evaluation (2024.lrec-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have made significant progress in recent years, but their practical use is hindered by their tendency to generate hallucinations. |
| Approach: | They propose to use ICD-10 and MeSH to evaluate LLMs' ability to detect medical hallucinations and make accurate diagnoses in noisy environments. |
| Outcome: | The proposed benchmark can be used to evaluate LLMs’ ability to detect medical hallucinations, make accurate diagnoses in noisy conditions, and provide plausible explanations. |
Revisit Self-Debugging with Self-Generated Tests for Code Generation (2025.acl-long)
Copied to clipboard
Xiancai Chen, Zhengwei Tao, Kechi Zhang, Changzhi Zhou, Xinyu Zhang, Wanli Gu, Yuanpeng He, Mengdi Zhang, Xunliang Cai, Haiyan Zhao, Zhi Jin
| Challenge: | Large language models (LLMs) have made significant advances in code generation, but they still face challenges when tackling complex programming tasks beyond their basic capabilities. |
| Approach: | They propose to integrate self-generated tests into the code generation process . they propose to use post-execution and in-exection self-debugging to mitigate test bias . |
| Outcome: | The proposed method improves the performance of large language models in code generation tasks by leveraging execution feedback from tests. |
Integrating Physician Diagnostic Logic into Large Language Models: Preference Learning from Process Feedback (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing studies have shown that large language models can enhance response richness and coherence, but there is a pressing need to bolster the model’s capacity for diagnostic logic to ensure patient safety. |
| Approach: | They propose an approach termed preference learning from process feedback (PLPF) that integrates the doctor’s diagnostic logic into LLMs. |
| Outcome: | The proposed approach improves the diagnostic accuracy of the baseline model in medical conversations by 17.6%, surpassing the performance of traditional approaches. |
MALoRA: Mixture of Asymmetric Low-Rank Adaptation for Enhanced Multi-Task Learning (2025.findings-naacl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) can be fine-tuned to new tasks, but in multi-task scenarios, training imbalance and seesaw effect often arise. |
| Approach: | They propose a flexible fine-tuning framework that leverages asymmetric optimization among LoRA experts to reduce training imbalance and improve performance. |
| Outcome: | The proposed framework outperforms baseline methods in inter- and intra-task learning scenarios. |
Beyond Input Activations: Identifying Influential Latents by Gradient Sparse Autoencoders (2025.emnlp-main)
Copied to clipboard
| Challenge: | Sparse Autoencoders (SAEs) have recently emerged as powerful tools for interpreting and steering the internal representations of large language models (LLMs). |
| Approach: | They propose a method that identifies the most influential latents by incorporating output-side gradient information. |
| Outcome: | The proposed method identifies the most influential latents by incorporating output-side gradient information. |
SEAG: Structure-Aware Event Causality Generation (2023.findings-acl)
Copied to clipboard
Zhengwei Tao, Zhi Jin, Xiaoying Bai, Haiyan Zhao, Chengfeng Dou, Yongqiang Zhao, Fang Wang, Chongyang Tao
| Challenge: | Current methods for extracting event causality are limited by the lack of cross-task dependencies and may cause error propagation. |
| Approach: | They propose an approach for Structure-Aware Event Causality Generation (SEAG) they generate the ECG structure using a pre-trained language model and perform structural discriminative training alongside auto-regressive generation. |
| Outcome: | The proposed method is effective in extracting event causality from text. |
Denoising Concept Vectors with Sparse Autoencoders for Improved Language Model Steering (2026.findings-eacl)
Copied to clipboard
| Challenge: | Existing methods for steering concept vectors suffer from noisy features in diverse datasets that undermine steering robustness. |
| Approach: | They propose a Sparse Autoencoder-Denoised Concept Vector (SDCV) which selectively keeps the most discriminative SAE latents while reconstructing hidden representations. |
| Outcome: | The proposed method improves steering success rates by 4-16% across six challenging concepts while maintaining topic relevance. |
UniEvent: Unified Generative Model with Multi-Dimensional Prefix for Zero-Shot Event-Relational Reasoning (2023.acl-long)
Copied to clipboard
| Challenge: | Reasoning about events and their relations is an indispensable ability to fulfill various event-centric or common-sense reasoning tasks. |
| Approach: | They propose a multi-task learning framework that organizes event relational reasoning tasks into a coordinate system with multiple axes, representing inter-event relations and reasoning formulations. |
| Outcome: | The proposed framework achieves state-of-the-art or competitive performance on zero-shot and supervised reasoning tasks. |
PlugMed: Improving Specificity in Patient-Centered Medical Dialogue Generation using In-Context Learning (2023.findings-emnlp)
Copied to clipboard
| Challenge: | In-context learning is a key task in health conversational assistants, but it is difficult to guarantee the specificity of the responses. |
| Approach: | They propose a plug-and-play medical dialogue system that provides a patient-centered medical interpretation service to users who are less knowledgeable about medical knowledge. |
| Outcome: | The proposed model improves the specificity of the patient-centered medical dialogues by providing them with real dialogues from similar patients as prompts. |