Papers by Rui Pan
MLLM-Protector: Ensuring MLLM’s Safety without Hurting Performance (2024.emnlp-main)
Copied to clipboard
Renjie Pi, Tianyang Han, Jianshu Zhang, Yueqi Xie, Rui Pan, Qing Lian, Hanze Dong, Jipeng Zhang, Tong Zhang
| Challenge: | MLLMs are deployed on limited image-text pairs, which makes them more vulnerable to catastrophic forgetting of their original abilities during safety fine-tuning. |
| Approach: | They propose a plug-and-play strategy that detects harmful visual inputs and transforms harmful ones into harmless ones. |
| Outcome: | The proposed approach mitigates the risks posed by malicious visual inputs without compromising the original performance of MLLMs. |
AppBench: Planning of Multiple APIs from Various APPs for Complex User Instruction (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing state-of-the-art Large Language Models (LLMs) still cannot perform well in this situation even with the help of in-context learning and finetuning. |
| Approach: | They propose a benchmark to evaluate LLMs’ ability to plan and execute multiple APIs from various sources in order to complete the user’s task. |
| Outcome: | The proposed benchmarks show that the existing state-of-the-art LLMs still cannot perform well in this situation even with in-context learning and finetuning. |
BlindGuard: Safeguarding LLM-based Multi-Agent Systems under Unknown Attacks (2026.acl-long)
Copied to clipboard
| Challenge: | Existing supervised defense methods rely on labeled malicious agents to train a supervised model of malicious behavior. |
| Approach: | They propose an unsupervised defense method that learns without requiring any attack-specific labels or prior knowledge of malicious behaviors. |
| Outcome: | The proposed method detects diverse attack types across MAS with various communication patterns while maintaining superior generalizability compared to baselines. |
DetGPT: Detect What You Need via Reasoning (2023.emnlp-main)
Copied to clipboard
Renjie Pi, Jiahui Gao, Shizhe Diao, Rui Pan, Hanze Dong, Jipeng Zhang, Lewei Yao, Jianhua Han, Hang Xu, Lingpeng Kong, Tong Zhang
| Challenge: | Recent advances in the field of computer vision have enabled more effective and sophisticated interactions between humans and machines. |
| Approach: | They propose a reasoning-based object detection paradigm that leverages state-of-the-art multi-modal models and open-vocabulary object detectors to perform reasoning within the context of the user’s instructions and the visual scene. |
| Outcome: | The proposed method enables users to interact with the system using natural language instructions, allowing for a higher level of interactivity. |
The Instinctive Bias: Spurious Images lead to Illusion in MLLMs (2024.emnlp-main)
Copied to clipboard
| Challenge: | Existing multi-modal large language models (MLLMs) are able to process visual inputs by converting them into visual tokens that share the same latent space as language tokens in LLMs. |
| Approach: | They propose a benchmark that assesses the visual illusion level given spurious images and a pipeline that converts visual inputs into visual tokens. |
| Outcome: | The proposed benchmark shows that MLLMs suffer from an instinctive bias to varying degrees when presented with spurious images. |
SongComposer: A Large Language Model for Lyric and Melody Generation in Song Composition (2025.acl-long)
Copied to clipboard
Shuangrui Ding, Zihan Liu, Xiaoyi Dong, Pan Zhang, Rui Qian, Junhao Huang, Conghui He, Dahua Lin, Jiaqi Wang
| Challenge: | Creating lyrics and melodies in symbolic format requires expert knowledge of melody and an advanced understanding of lyrics. |
| Approach: | They introduce SongComposer, a music-specialized large language model that can create symbolic lyrics and melodies following instructions. |
| Outcome: | The proposed model outperforms existing models in symbolic song composition tasks. |
Mitigating the Alignment Tax of RLHF (2024.emnlp-main)
Copied to clipboard
Yong Lin, Hangyu Lin, Wei Xiong, Shizhe Diao, Jianmeng Liu, Jipeng Zhang, Rui Pan, Haoxiang Wang, Wenbin Hu, Hanning Zhang, Hanze Dong, Renjie Pi, Han Zhao, Nan Jiang, Heng Ji, Yuan Yao, Tong Zhang
| Challenge: | Large Language Models (LLMs) acquire a wide range of abilities during pre-training, but aligning LLMs under Reinforcement Learning with Human Feedback (RLHF) can lead to forgetting pretrained abilities, which is also known as the alignment tax. |
| Approach: | They propose to use a model averaging technique to find the most powerful alignment-forging Pareto front among RLHF algorithms. |
| Outcome: | The proposed method achieves the strongest alignment-forging Pareto front among competing methods. |
DART: Open-Domain Structured Data Record to Text Generation (2021.naacl-main)
Copied to clipboard
Linyong Nan, Dragomir Radev, Rui Zhang, Amrit Rau, Abhinand Sivaprasad, Chiachun Hsieh, Xiangru Tang, Aadit Vyas, Neha Verma, Pranav Krishna, Yangxiaokang Liu, Nadia Irwanto, Jessica Pan, Faiaz Rahman, Ahmad Zaidi, Mutethia Mutuma, Yasin Tarabar, Ankit Gupta, Tao Yu, Yi Chern Tan, Xi Victoria Lin, Caiming Xiong, Richard Socher, Nazneen Fatema Rajani
| Challenge: | Data-to-text annotations can be costly when dealing with tables with nontrivial structures. |
| Approach: | They propose a procedure for extracting semantic triples from tables that encodes their structures by exploiting table headers and table title. |
| Outcome: | The proposed method exploits the semantic dependencies between table headers and title to extract semantic triples from tables. |
GUIDE: Towards Scalable Advising for Research Ideas (2026.acl-long)
Copied to clipboard
| Challenge: | Existing systems that provide detailed, constructive feedback on academic papers struggle with review fidelity. |
| Approach: | They explore factors that underlie the development of robust advising systems . large language models have shown remarkable progress in tasks from text generation to code synthesis . |
| Outcome: | The proposed model outperforms general-purpose language models in acceptance rates for self-ranked top-30% submissions to ICLR 2025. |
Active Prompting with Chain-of-Thought for Large Language Models (2024.acl-long)
Copied to clipboard
| Challenge: | Existing methods to annotate large language models rely on a fixed set of human-annotated exemplars, which are not always the most effective for different tasks. |
| Approach: | They propose a method to adapt large language models to different tasks with task-specific example prompts (annotated with human-designed CoT reasoning) they introduce several metrics to characterize uncertainty so as to select the most uncertain questions for annotation. |
| Outcome: | The proposed method significantly improves performance on eight complex reasoning tasks. |
MiCRo: Mixture Modeling and Context-aware Routing for Personalized Preference Learning (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing reward models assume a global reward function, limiting personalization and pluralistic alignment. |
| Approach: | They propose a framework that leverages binary preference datasets to enhance personalized preference learning. |
| Outcome: | The proposed framework captures diverse human preferences without fine-grained annotations and significantly improves personalized preference learning on downstream tasks. |
Plum: Prompt Learning using Metaheuristics (2024.findings-acl)
Copied to clipboard
Rui Pan, Shuo Xing, Shizhe Diao, Wenhe Sun, Xiang Liu, KaShun Shum, Jipeng Zhang, Renjie Pi, Tong Zhang
| Challenge: | Recent advances in prompt learning have led to a need for general prompt optimization methods. |
| Approach: | They propose a branch of discrete non-convex optimization methods with over 100 options as a promising approach to prompt learning. |
| Outcome: | The proposed methods can be used to discover more human-understandable prompts that were previously unknown in reasoning and image generation tasks. |
Mitigating Context Interference for Reliable and Efficient Search Agents (2026.acl-long)
Copied to clipboard
Boyang Xue, Bin Wu, Shuofei Qiao, Sheng Wang, Rui Wang, Yiming Du, Hongru Wang, Jeff Z. Pan, Emine Yilmaz, Kam-Fai Wong, Aldo Lipani
| Challenge: | Recent research empowers Large Language Models (LLMs) as multi-turn search agents to iteratively retrieve and generate outputs until complex tasks are solved. |
| Approach: | They propose a distill-based context refiner to dynamically mitigate context interference . they also propose RLs that refine contexts to generate outputs . |
| Outcome: | The proposed refiner can mitigate context interference in multi-turn search agents. |
AntiLeakBench: Preventing Data Contamination by Automatically Constructing Benchmarks with Updated Real-World Knowledge (2025.acl-long)
Copied to clipboard
Xiaobao Wu, Liangming Pan, Yuxi Xie, Ruiwen Zhou, Shuai Zhao, Yubo Ma, Mingzhe Du, Rui Mao, Anh Tuan Luu, William Yang Wang
| Challenge: | Existing studies solve this challenge by updating benchmarks with newly collected data, but they fail to guarantee contamination-free evaluation as the newly collected knowledge may contain pre-existing knowledge. |
| Approach: | They propose an automated anti-leakage benchmarking framework that builds and updates benchmarks without human labor instead of using newly collected data. |
| Outcome: | The proposed framework significantly reduces the cost of benchmark maintenance to accommodate emerging LLMs. |
Comprehensive Benchmarking of Long-Form Speech Generation in Diverse Scenarios (2026.findings-acl)
Copied to clipboard
Changhao Pan, Rui Yang, Han Wang, Zhuan Zhou, Xuming He, Wenxiang Guo, Ziyue Jiang, Ruiqi Li, Yu Zhang, Chenyuhao Wen, Ke Lei, Xiang Yin, Jingyu Lu, Zhiyuan Zhu, Zhou Zhao
| Challenge: | Existing evaluation benchmarks for long-form speech are limited to limited domains, creating a significant gap with the diverse downstream applications. |
| Approach: | They propose a benchmark that decomposes "long-form speech quality" into specific, disentangled dimensions. |
| Outcome: | The proposed benchmark decomposes “long-form speech quality” into specific, disentangled dimensions. |
TopoDIM: One-shot Topology Generation of Diverse Interaction Modes for Multi-Agent Systems (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing communication topologies rely on spatio-temporal dialogues, which incur high latency and computation. |
| Approach: | They propose a framework for one-shot Topology generation with Diverse Interaction Modes that enables agents to construct heterogeneous communication without iterative coordination. |
| Outcome: | The proposed framework reduces total token consumption by 46.41% while improving average performance by 1.50% over state-of-the-art methods. |
TheoremLlama: Transforming General-Purpose LLMs into Lean4 Experts (2024.emnlp-main)
Copied to clipboard
| Challenge: | a framework for formal proof writing using formal languages like Lean4 is needed to prove mathematical theorems using formal language. |
| Approach: | They propose a framework that trains a general-purpose LLM to be a Lean4 expert. |
| Outcome: | The proposed framework achieves cumulative accuracies of 36.48% and 33.61% on MiniF2F-Valid and Test datasets. |
LMFlow: An Extensible Toolkit for Finetuning and Inference of Large Foundation Models (2024.naacl-demo)
Copied to clipboard
| Challenge: | Foundation models have demonstrated a great ability to achieve general human-level intelligence far beyond traditional approaches. |
| Approach: | They propose a toolkit to simplify the finetuning of general foundation models. |
| Outcome: | The proposed toolkit simplifies the domain- and task-aware finetuning of general foundation models with limited computing resources. |
Bridge-Coder: Transferring Model Capabilities from High-Resource to Low-Resource Programming Language (2025.findings-acl)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) excel at generating code for high-resource programming languages (HRPLs) however, they struggle significantly with low-resourced programming languages such as D, exacerbating the digital divide. |
| Approach: | They propose a method to generate LRPL data using LLM's general knowledge, HRPL proficiency, and in-context learning capabilities. |
| Outcome: | The proposed method improves on R, D, Racket, and Bash, while maintaining the same quality. |
TAGCOS: Task-agnostic Gradient Clustered Coreset Selection for Instruction Tuning Data (2025.findings-naacl)
Copied to clipboard
| Challenge: | Existing methods for instruction tuning are limited due to the increasing volume of instruction datasets and the increased computational costs. |
| Approach: | They propose to extract a small and highly informative subset of training samples from a large dataset that achieves comparable performance to the full dataset. |
| Outcome: | The proposed algorithm outperforms other unsupervised methods and achieves comparable performance to the full dataset. |
Explainable and Fine-Grained Safeguarding of LLM Multi-Agent Systems via Bi-Level Graph Anomaly Detection (2026.acl-long)
Copied to clipboard
Junjun Pan, Yixin Liu, Rui Miao, Kaize Ding, Yu Zheng, Quoc Viet Hung Nguyen, Alan Wee-Chung Liew, Shirui Pan
| Challenge: | Existing graph anomaly detection methods rely on coarse sentence-level information and overlook fine-grained lexical cues, limiting their reliability and real-world applicability. |
| Approach: | They propose an explainable and fine-grained safeguarding framework for detecting malicious agents in multi-agent systems (MAS) to incorporate both coarse and fine lexical information for anomalous agent identification. |
| Outcome: | Extensive experiments across diverse MAS topologies and attack scenarios demonstrate robust detection performance and strong interpretability of XG-Guard. |
MathFusion: Enhancing Mathematical Problem-solving of LLM through Instruction Fusion (2025.acl-long)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have shown impressive progress in mathematical problem-solving . current approaches to enhance mathematical reasoning focus on instance-level modifications . |
| Approach: | They propose a framework that enhances mathematical reasoning through cross-problem instruction synthesis. |
| Outcome: | The proposed framework boosts mathematical reasoning by 18.0 points while maintaining high data efficiency. |
ScaleBiO: Scalable Bilevel Optimization for LLM Data Reweighting (2025.acl-long)
Copied to clipboard
Rui Pan, Dylan Zhang, Hanning Zhang, Xingyuan Pan, Minrui Xu, Jipeng Zhang, Renjie Pi, Xiaoyu Wang, Tong Zhang
| Challenge: | Existing paradigms for bilevel optimization require second-order information, making it difficult to scale them up. |
| Approach: | They propose a scalable instantiation of a bilevel optimization paradigm for large-scale LLMs by using a memory-efficient training technique. |
| Outcome: | The proposed paradigm scales to 30B-sized LLMs on 8H100 GPUs. |
Feeling Rules in Language Models: Mapping Norms of Emotional Appropriateness Across Roles, Institutions, and Intensity (2026.acl-long)
Copied to clipboard
| Challenge: | Existing benchmarks measure whether Large language models recognize emotions . authors: LLMs can be used to validate, but they can still judge anger inappropriately . |
| Approach: | They propose a benchmark to measure whether Large language models validate anger . they use explicit norm judgments and implicit acceptability tests to measure norms . |
| Outcome: | The study finds that large differences in sanctioning thresholds and institutional norm signatures are not reducible to overall strictness. |
SpiderFlow: Efficient Topology-Aware Scheduling for LLM Training Across Decentralized GPU Clusters (2026.acl-long)
Copied to clipboard
| Challenge: | Existing approaches to training large language models lack topologyaware task scheduling mechanisms and model parallelization strategies. |
| Approach: | They propose a topology-aware scheduling system specifically designed for decentralized GPU clusters . they propose heuristic methods at the inter-cluster level with ILP-based optimization within clusters. |
| Outcome: | The proposed system reduces job completion time by 1.2-1.3 and improves throughput by 1.12-1.25 . it also reduces scheduling overhead by 20-90 on average compared to state-of-the-art scheduling systems. |
Self-Sum: Teaching an Agent to Decide Itself When and What to Summarize (2026.findings-acl)
Copied to clipboard
Hongru Wang, Rui Wang, Jushi Kai, Boyang Xue, Yongqi Li, Shijue Huang, Xiaoteng Ma, Jeff Z. Pan, Amos Storkey
| Challenge: | Existing methods for summarizing long-horizon agents rely on fixed, rule-based summarization strategies. |
| Approach: | They propose a framework that empowers agents to autonomously decide when and what to summarize by modeling it as an internal cognitive action unified with environmental actions. |
| Outcome: | The proposed framework outperforms no-summarization and rule-based training methods on long-horizon benchmarks and shows strong generalization gains. |
Multi-Task Neural Model for Agglutinative Language Translation (2020.acl-srw)
Copied to clipboard
| Challenge: | Neural machine translation (NMT) has been gaining popularity in high-resource translation tasks, but struggles in low-ressource and morphologically-rich scenarios. |
| Approach: | They propose a multi-task neural model that jointly learns to perform bi-directional translation and agglutinative language stemming. |
| Outcome: | The proposed model can significantly improve translation performance on agglutinative languages by using a small amount of monolingual data. |
LexGenius: An Expert-Level Benchmark for Large Language Models in Legal General Intelligence (2026.findings-acl)
Copied to clipboard
Wenjin Liu, Haoran Luo, Xin Feng, Xiang Ji, Lijuan Zhou, Rui Mao, Jiapu Wang, Shirui Pan, Erik Cambria
| Challenge: | Existing benchmarks for legal general intelligence (GI) are result-oriented and do not evaluate the legal intelligence of large language models (LLMs). |
| Approach: | They propose a Chinese legal benchmark for evaluating legal GI in large language models . they use recent legal cases and exam questions to create multiple-choice questions . |
| Outcome: | The proposed benchmarks lack a systematic evaluation of the legal intelligence of large language models (LLMs) the results show that even the best LLMs lagging behind human legal professionals. |
Grounding Visual Illusions in Language: Do Vision-Language Models Perceive Illusions Like Humans? (2023.emnlp-main)
Copied to clipboard
| Challenge: | Visual illusions are a phenomenon that is often seen in human perception but are not always faithful to the physical world. |
| Approach: | They build a dataset containing five types of visual illusions and formulate four tasks to examine visual illusion in state-of-the-art VLMs. |
| Outcome: | The proposed dataset reveals that larger models are closer to human perception and more susceptible to visual illusions. |
Understanding the Information Propagation Effects of Communication Topologies in LLM-based Multi-Agent Systems (2025.emnlp-main)
Copied to clipboard
| Challenge: | Empirical studies for communication topology design often overlook why and when sparse and dense topologies help or hinder collaboration. |
| Approach: | They propose a topology design approach that balances error suppression and beneficial information propagation by fusing connectivity patterns from dense and sparse graphs. |
| Outcome: | The proposed topology design achieves superior performance across tasks with sparse and dense graphs. |