Papers by Xinyu Yan
AFT-Tab: Adversarial Fine-Tuning for Tabular Data Synthesis with Long Text Columns (2026.acl-long)
Copied to clipboard
| Challenge: | Existing tabular data synthesis methods fail to account for cross-modal heterogeneity of real-world tables, where structured continuous and discrete attributes coexist with unstructured long-text columns. |
| Approach: | They propose a framework that synergistically trains an LLM-based text generator and a deep-learning-based non-textual generator to quantify cross-modal semantic alignment. |
| Outcome: | The proposed framework outperforms state-of-the-art frameworks in fidelity, diversity, and task utility. |
Towards Robustness of Text-to-SQL Models Against Natural and Realistic Adversarial Table Perturbation (2022.acl-long)
Copied to clipboard
| Challenge: | Existing Text-to-SQL parsers are vulnerable to perturbations in NL questions . we propose the Adversarial Table Perturbation (ATP) as a new attacking paradigm . |
| Approach: | They propose to use the Adversarial Table Perturbation to measure robustness of Text-to-SQL parsers against adversarial perturbations. |
| Outcome: | The proposed approach outperforms baseline methods in robustness evaluations on ADVETA and can be used in future projects. |
Reasoning Like Program Executors (2022.emnlp-main)
Copied to clipboard
Xinyu Pi, Qian Liu, Bei Chen, Morteza Ziyadi, Zeqi Lin, Qiang Fu, Yan Gao, Jian-Guang Lou, Weizhu Chen
| Challenge: | Existing language models are inadequate in reasoning, according to studies . a new reasoning pre-training paradigm is based on pretraining language models with programs . |
| Approach: | They propose a reasoning pre-training paradigm that empowers language models to harvest reasoning knowledge possessed by program executors. |
| Outcome: | The proposed reasoning pre-training paradigm can boost models' reasoning skills . it can be instantiated by different kinds of program executors and run on a single database . |
RealBench: A Chinese Multi-image Understanding Benchmark Close to Real-world Scenarios (2025.findings-emnlp)
Copied to clipboard
Fei Zhao, Chengqiang Lu, Yufan Shen, Qimeng Wang, Yicheng Qian, Haoxin Zhang, Yan Gao, null Yiwu, Yao Hu, Zhen Wu, Shangyu Xing, Xinyu Dai
| Challenge: | RealBench is the first Chinese multimodal multi-image dataset . the dataset contains 9393 samples and 69910 images . |
| Approach: | They propose to create a Chinese multimodal multi-image dataset using 21 models . they use closed-source models that support multi-inputs as well as open-source visual and video models a . |
| Outcome: | The first Chinese multimodal multi-image dataset contains 9393 samples and 69910 images. |
STORM-BORN: A Challenging Mathematical Derivations Dataset Curated via a Human-in-the-Loop Multi-Agent Framework (2025.findings-acl)
Copied to clipboard
Wenhao Liu, Zhenyi Lu, Xinyu Hu, Jerry Zhang, Dailin Li, Jiacheng Cen, Huilin Cao, Haiteng Wang, Yuhan Li, Xie Kun, Dandan Li, Pei Zhang, Chengbo Zhang, Yuxiang Ren, Xiaohong Huang, Yan Ma
| Challenge: | Existing datasets suffer from outdated and insufficient challenging content, neglecting human-like reasoning, and limited reliability due to single-LLM generation. |
| Approach: | They propose a human-in-the-loop, multi-agent data generation framework that integrates reasoning-dense filters, multiagent collaboration, and human mathematicians’ evaluations to ensure the reliability and quality of the dataset. |
| Outcome: | The proposed framework improves accuracy and quality of the 2,000-synthesized datasets by integrating reasoning-dense filters, multi-agent collaboration, and human mathematicians’ evaluations. |
Is ChatGPT Good at Search? Investigating Large Language Models as Re-Ranking Agents (2023.emnlp-main)
Copied to clipboard
Weiwei Sun, Lingyong Yan, Xinyu Ma, Shuaiqiang Wang, Pengjie Ren, Zhumin Chen, Dawei Yin, Zhaochun Ren
| Challenge: | Existing work utilizes generative LLMs for Information Retrieval (IR) rather than direct passage ranking. |
| Approach: | They investigate generative LLMs such as ChatGPT and GPT-4 for relevance ranking in IR and use a test set to verify the model’s ability to rank unknown knowledge. |
| Outcome: | The proposed model outperforms a 3B supervised model on the BEIR benchmark. |
Multi-Hop Knowledge Editing via Critic-Guided Multi-Agent Reasoning (2026.findings-acl)
Copied to clipboard
Xudong Li, Yuhang Tian, Dandan Song, Zhijing Wu, Shuhao Zhang, Jun Yang, Yongyu Huo, Changzhi Zhou, Xinyu Zhang, Chenhao Li, Huipeng Ma, Luan Zhang, Yan Xu, Qian Liu
| Challenge: | Existing knowledge editing methods rely on unidirectional, feed-forward pipelines . a minor retrieval error or logical mismatch at an early hop can become a silent failure . |
| Approach: | They propose a framework for closed-loop post-edit reasoning that uses a Critic agent to verify coherence and step-wise correctness. |
| Outcome: | Experiments on MQuAKE-2002 and MQuADE-hard show that CARE effectively mitigates error propagation . a minor retrieval error or logical mismatch at an early hop can become a silent failure . |
Structural Knowledge Distillation: Tractably Distilling Information for Structured Predictor (2021.acl-long)
Copied to clipboard
Xinyu Wang, Yong Jiang, Zhaohui Yan, Zixia Jia, Nguyen Bach, Tao Wang, Zhongqiang Huang, Fei Huang, Kewei Tu
| Challenge: | Knowledge distillation is a technique to transfer knowledge between models, typically from a large model (the teacher) to a more fine-grained one (the student). |
| Approach: | They propose a factorized form of the knowledge distillation objective for structured prediction which is tractable for many typical choices of the teacher and student models. |
| Outcome: | The proposed model is able to transfer knowledge between teacher and student models without loss of accuracy under four different scenarios. |
Demystify the Role of Memory in Machine Learning Engineering Agents (2026.findings-acl)
Copied to clipboard
Xinyu Zhao, Junpeng Wang, Yuzhong Chen, Menghai Pan, Chin-Chia Michael Yeh, Jiarui Sun, Yan Zheng, Mahashweta Das, Tianlong Chen
| Challenge: | Unlike short, reactive exchanges, MLE agents solve tasks through cycles of experimentation and improvement where past errors can inform future success. |
| Approach: | They propose a dynamic coding memory that captures and reuses debugging experiences and integrates it into two representative agent paradigms. |
| Outcome: | The proposed agent model captures and reuses debugging experiences and integrates it into two agent paradigms. |
Mirror: Multiple-perspective Self-Reflection Method for Knowledge-rich Reasoning (2024.acl-long)
Copied to clipboard
| Challenge: | Large language models (LLMs) struggle with knowledge-rich problems without external resources. |
| Approach: | They propose a Multiple-perspective self-reflection method that allows LLMs to reflect from multiple-perceptive clues, achieved through a heuristic interaction between a Navigator and a Reasoner. |
| Outcome: | The proposed method is superior to other self-reflection methods on five reasoning datasets. |
SEMIROUTER: Sparse-Data Enhanced Routing for Adaptive Multi-LLM System (2026.eacl-long)
Copied to clipboard
| Challenge: | Existing routing methods suffer from poor scalability and dependence on datasets for training . energy footprint is also considered in the decision to implement our new LLM routing framework . |
| Approach: | They propose a new LLM routing framework that dynamically allocates queries to the most appropriate LLM. |
| Outcome: | The proposed method improves data efficiency, adaptability, and routing accuracy compared to existing methods. |
MAIR: A Massive Benchmark for Evaluating Instructed Retrieval (2024.emnlp-main)
Copied to clipboard
Weiwei Sun, Zhengliang Shi, Wu Long, Lingyong Yan, Xinyu Ma, Yiding Liu, Min Cao, Dawei Yin, Zhaochun Ren
| Challenge: | Existing IR benchmarks focus on a limited scope of tasks, making them insufficient for evaluating the latest IR models. |
| Approach: | They propose a multi-task instruction-tuned IR benchmark that includes 126 distinct IR tasks across 6 domains. |
| Outcome: | The proposed model performs better on instruction-tuned models than non-instruction-tunned models on MAIR. |