Papers by Xinyu Lu
EvoEdit: Evolving Null-space Alignment for Robust and Efficient Knowledge Editing (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to modifying large language models require continual updates to rectify outdated or erroneous knowledge. |
| Approach: | They propose a model editing strategy that mitigates catastrophic interference through sequential null-space alignment. |
| Outcome: | EvoEdit achieves better or comparable performance than prior state-of-the-art techniques with up to 3.53 speedup. |
ToM-Synth: Scaling Robust Theory of Mind in LLMs via 6,912 Structured Social Units (2026.findings-acl)
Copied to clipboard
Guiyang Hou, Xiang Huang, Shangke Lyu, Yuchuan Wu, Weiyao Luo, Xinyu Mei, Yongliang Shen, Weiming Lu, Yongbin Li
| Challenge: | Existing methods endowing LLMs with Theory of Mind fail to internalize the augmented ToM into the LLM. |
| Approach: | They propose a factorial combinatorial synthesis framework that enables systematic synthesis of ToM data and uses it for RL fine-tuning. |
| Outcome: | The proposed framework yields a training dataset of 27,648 instances. |
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| Challenge: | Existing datasets suffer from outdated and insufficient challenging content, neglecting human-like reasoning, and limited reliability due to single-LLM generation.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. |
An Unsupervised Method for Building Sentence Simplification Corpora in Multiple Languages (2021.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods to build parallel sentence simplification corpora are limited . SS is used to rephrase sentences into simpler forms for those with cognitive disabilities . |
| Approach: | They propose to build SS corpora from large-scale bilingual translation corpors using a parallel approach. |
| Outcome: | The proposed method outperforms the existing methods on WikiLarge and achieves state-of-the-art results. |
On-Policy Self-Alignment with Fine-grained Knowledge Feedback for Hallucination Mitigation (2025.findings-acl)
Copied to clipboard
Xueru Wen, Jie Lou, Xinyu Lu, Yuqiu Ji, Xinyan Guan, Yaojie Lu, Hongyu Lin, Ben He, Xianpei Han, Debing Zhang, Le Sun
| Challenge: | Large language models exhibit behavior that deviates from the boundaries of their knowledge during response generation. |
| Approach: | They propose a framework that allows large language models to explore their knowledge boundaries and self-correct generation behavior through fine-grained feedback signals. |
| Outcome: | The proposed framework enables LLMs to explore their knowledge boundaries and self-correct generation behavior through fine-grained feedback signals. |
MathCanvas: Intrinsic Visual Chain-of-Thought for Multimodal Mathematical Reasoning (2026.acl-long)
Copied to clipboard
Weikang Shi, Aldrich Yu, Rongyao Fang, Houxing Ren, Ke Wang, Aojun Zhou, Changyao Tian, Xinyu Fu, Yuxuan Hu, Zimu Lu, Linjiang Huang, Si Liu, Rui Liu, Hongsheng Li
| Challenge: | Existing approaches to visual chain-of-thought are limited by external tools or fail to generate high-fidelity diagrams. |
| Approach: | They propose a framework to enable large multimodal models with VCoT capabilities . they pre-train a model on a 15.2M-pair corpus and teach it how to leverage visual aids . |
| Outcome: | The proposed framework unlocks complex, human-like visual reasoning in large language models . it pre-trains the model on a 15.2M-pair corpus and fine-tunes it on MathCanvas-Instruct . |
Improving Knowledge Graph Completion with Structure-Aware Supervised Contrastive Learning (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing contrastive methods focus on individual triples, overlooking the broader structural connectivities and topologies of KGs. |
| Approach: | They propose a new contrastive learning framework that incorporates four tasks specifically tailored to KG data: Vertex-level CL, Neighbor-level Cl, Path-levelCL, and Relation composition level CL. |
| Outcome: | The proposed framework achieves SOTA performance under standard supervised and low-resource settings. |
Is the Attention Matrix Really the Key to Self-Attention in Multivariate Long-Term Time Series Forecasting? (2026.acl-long)
Copied to clipboard
| Challenge: | In multivariate long-term time series forecasting, it is widely believed that the effectiveness of self-attention arises from its attention matrix. |
| Approach: | They propose a multi-branch MLP that isolates the ‘multi-brain mapping with element-wise operation’ structure from the Transformer and shows that it achieves competitive performance. |
| Outcome: | The proposed model outperforms three classic and three latest Transformer models and shows that it achieves competitive performance. |
Named Entity and Relation Extraction with Multi-Modal Retrieval (2022.findings-emnlp)
Copied to clipboard
| Challenge: | Existing approaches to name entity recognition and relation extraction are knowledge-based and may not be highly relevant. |
| Approach: | They propose a multi-modal named entity recognition framework that leverages image information to improve the performance of NER and relation extraction. |
| Outcome: | The proposed framework can achieve state-of-the-art on four multi-modal named entity recognition datasets and one multi-module relation extraction dataset. |
FaStFact: Faster, Stronger Long-Form Factuality Evaluations in LLMs (2025.findings-emnlp)
Copied to clipboard
Yingjia Wan, Haochen Tan, Xiao Zhu, Xinyu Zhou, Zhiwei Li, Qingsong Lv, Changxuan Sun, Jiaqi Zeng, Yi Xu, Jianqiao Lu, Yinhong Liu, Zhijiang Guo
| Challenge: | Prior evaluation pipelines fail to evaluate factuality of long-form LLMs due to inefficiency and costly human assessment. |
| Approach: | They propose a fast and strong evaluation pipeline that can evaluate factuality of long-form LLMs . they propose 'faStFact' to reduce cost of web searching and inference calling . |
| Outcome: | The proposed evaluation pipeline achieves highest alignment with human evaluation and efficiency among existing baselines. |
Debiasing In-Context Learning by Instructing LLMs How to Follow Demonstrations (2024.findings-acl)
Copied to clipboard
Lvxue Li, Jiaqi Chen, Xinyu Lu, Yaojie Lu, Hongyu Lin, Shuheng Zhou, Huijia Zhu, Weiqiang Wang, Zhongyi Liu, Xianpei Han, Le Sun
| Challenge: | In-context learning (ICL) has gained considerable attention due to its data efficiency and task adaptability. |
| Approach: | They propose to de-biase demonstration bias in in-context learning by focusing on semantic ambiguity induced by demonstrations and reducing the semantic hazard. |
| Outcome: | The proposed methods significantly improve performance on six datasets. |
Transferable Post-training via Inverse Value Learning (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing algorithms for post-training large datasets are requiring a large computational effort. |
| Approach: | They propose to model the changes at logits level during post-training using a separate neural network . they demonstrate that the value network can be seamlessly integrated with another pre-trained model . |
| Outcome: | The proposed model can be integrated with another pre-trained model during inference, enabling similar capability enhancements. |
AutoAlign: Get Your LLM Aligned with Minimal Annotations (2025.acl-demo)
Copied to clipboard
Xinyu Lu, Dong Xu, Chunkang Zhang, Xinyan Guan, Junxiang Wang, Qingyu Zhang, Pengbo Wang, Yingzhi Mao, Hao Xiang, Xueru Wen, Zichao Li, Yaojie Lu, Hongyu Lin, Le Sun, Xianpei Han
| Challenge: | Automated Alignment (ALM) is a set of algorithms designed to align Large Language Models (LLMs) with human intentions and values while minimizing manual intervention. |
| Approach: | They propose an open-source toolkit that integrates mainstream automated algorithms through a consistent interface and an accessible workflow supporting one-click execution for prompt synthesis and automatic alignment signal construction. |
| Outcome: | The proposed framework enables easy reproduction of existing results through extensive benchmarks and facilitates the development of novel approaches via modular components. |
Argument Generation with Retrieval, Planning, and Realization (P19-1)
Copied to clipboard
| Challenge: | a novel argument generation framework is used to generate counter-arguments . CANDELA uses a text planning decoder to retrieve arguments of different perspectives . |
| Approach: | They propose a powerful retrieval system and a novel two-step argument generation framework . they use a retrieval-based retrieval platform indexed with 12 million articles from Wikipedia . |
| Outcome: | The proposed framework yields higher BLEU, ROUGE, and METEOR scores than state-of-the-art models. |
STRICT: Stress-Test of Rendering Image Containing Text (2025.emnlp-main)
Copied to clipboard
| Challenge: | Despite the advances in diffusion models, the generation of coherent text remains a major bottleneck. |
| Approach: | They propose a benchmark to test the ability of diffusion models to render coherent text in images. |
| Outcome: | The proposed model fails to generate coherent and legible text in images despite its iterative nature . the model fails in both the maximum length of readable text and correctness and legibility of the generated text . |
PAIR: Planning and Iterative Refinement in Pre-trained Transformers for Long Text Generation (2020.emnlp-main)
Copied to clipboard
| Challenge: | We present a content-controlled text generation framework for pre-trained Transformers . large pre-train models are the cornerstone of many state-of-the-art models in natural language understanding and generation tasks. |
| Approach: | They propose a content-controlled text generation framework that adds content planning to large pre-trained Transformers without modifying model architecture. |
| Outcome: | The proposed framework improves the quality of the outputs on three domains. |
MemTR: Enhancing Tool-Calling Reliability via Uncertainty-Triggered FFN-Space Retracing (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing tool-calling methods rely on costly tool-use training data or only constrain syntax, leaving tool selection and argument value errors largely unsolved. |
| Approach: | They propose a method that decodes tool evidence from the tool library and mixes it into the output at the uncertain layer. |
| Outcome: | The proposed method reduces tool calling failures by 2%–9% with only 1%–2% runtime overhead. |
Do Vision-Language Models Have Internal World Models? Towards an Atomic Evaluation (2025.findings-acl)
Copied to clipboard
Qiyue Gao, Xinyu Pi, Kevin Liu, Junrong Chen, Ruolan Yang, Xinqi Huang, Xinyu Fang, Lu Sun, Gautham Kishore, Bo Ai, Stone Tao, Mengyang Liu, Jiaxi Yang, Chao-Jung Lai, Chuanyang Jin, Jiannan Xiang, Benhao Huang, Zeming Chen, David Danks, Hao Su, Tianmin Shu, Ziqiao Ma, Lianhui Qin, Zhiting Hu
| Challenge: | Recent studies have evaluated and shown limitations in specific capabilities such as visual understanding, but a systematic evaluation of VLMs’ fundamental WM abilities remains absent. |
| Approach: | They propose a framework that assesses perception and prediction to provide an atomic evaluation of VLMs as WMs. |
| Outcome: | The proposed framework assesses perception and prediction abilities on 15 latest VLMs and compares them to human-level models. |
Let LLMs Take on the Latest Challenges! A Chinese Dynamic Question Answering Benchmark (2025.coling-main)
Copied to clipboard
Zhikun Xu, Yinghui Li, Ruixue Ding, Xinyu Wang, Boli Chen, Yong Jiang, Haitao Zheng, Wenlian Lu, Pengjun Xie, Fei Huang
| Challenge: | Recent work has noted that due to the extremely high cost of iterative updates of LLMs, they are often unable to answer dynamic questions well. |
| Approach: | They propose a Chinese Dynamic QA benchmark containing question-answer pairs related to the latest dynamic questions on the Chinese Internet. |
| Outcome: | The proposed benchmark will be one of the key data resources for improving LLMs’ Chinese question-answering ability in the future. |
DYPLOC: Dynamic Planning of Content Using Mixed Language Models for Text Generation (2021.acl-long)
Copied to clipboard
| Challenge: | Existing neural generation models fall short of coherence, thus requiring efficient content planning. |
| Approach: | They propose a generation framework that conducts dynamic planning of content while generating the output based on a novel design of mixed language models. |
| Outcome: | The proposed model outperforms competing models on argument generation and writing articles using New York Times’ Opinion section. |
Argument Mining for Understanding Peer Reviews (N19-1)
Copied to clipboard
| Challenge: | In 2015 alone, approximately 63.4 million hours were spent on peer reviews. |
| Approach: | They propose to automatically detect argumentative propositions put forward by reviewers and their types by automatically detecting their types and types. |
| Outcome: | The proposed method detects (1) the argumentative propositions put forward by reviewers, and (2) their types (e.g., evaluating the work or making suggestions for improvement). |
Efficient Argument Structure Extraction with Transfer Learning and Active Learning (2022.findings-acl)
Copied to clipboard
| Challenge: | Identifying and understanding the argumentative discourse structure in text has been a critical task in argument mining. |
| Approach: | They propose a context-aware Transformer-based argument structure prediction model that outperforms models that rely on features or only encode limited contexts. |
| Outcome: | The proposed model outperforms models that rely on features or encode limited contexts on five domains and on peer reviews on five different domains. |
MATCH: Modulating Attention via In-Context Retrieval for Long-Context Transformers (2026.acl-long)
Copied to clipboard
Linrui Ma, Chun Hei Lo, Xinyu Wang, Peng Lu, Xihao Yuan, Hanting Chen, Kai Han, Xinghao Chen, Chengjun Zhan, Hanlin xu, Yichun Yin, Lifeng Shang, Feng Wen, Boxing Chen, Yufei Cui
| Challenge: | Existing approaches to improve efficiency often enforce rigid structural constraints such as local attention windows. |
| Approach: | They propose a framework that augments sparse-attention mechanisms with dynamically integrated in-context information through an efficient retrieval system. |
| Outcome: | Empirical results show that MATCH significantly improves the performance of sparse-attention models on synthetic and real-world natural-language tasks. |
Sentence-Level Content Planning and Style Specification for Neural Text Generation (D19-1)
Copied to clipboard
| Challenge: | Recent advances in text generation systems often produce incoherent and unfaithful outputs . a novel automated text generation system takes into account content selection, text planning, and surface realization. |
| Approach: | They propose an end-to-end trained two-step text generation model that considers sentence-level content planners and language styles. |
| Outcome: | The proposed model outperforms competing models in three domains with diverse topics and varying language styles. |
Neural Argument Generation Augmented with Externally Retrieved Evidence (P18-1)
Copied to clipboard
| Challenge: | Existing methods for generating arguments are limited to retrieval-based methods. |
| Approach: | They propose an encoder-decoder-based argument generation model enriched with externally retrieved evidence from Wikipedia. |
| Outcome: | The proposed model generates arguments with more topic-relevant content than current models based on automatic evaluation and human assessments on a large-scale dataset from reddit. |
SoFA: Shielded On-the-fly Alignment via Priority Rule Following (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing alignment methods fail to adapt to the diversity of preferences and regulatory standards. |
| Approach: | They propose a method for prioritizing rules over user instructions to minimize misalignments in Large Language Models. |
| Outcome: | The proposed approach minimizes misalignments and adapts smoothly to various unseen rules, ensuring they are shielded from hijacking and that the model responds appropriately. |