Papers by Yao Tong
A Mousetrap: Fooling Large Reasoning Models for Jailbreak with Chain of Iterative Chaos (2025.findings-acl)
Copied to clipboard
| Challenge: | Large Reasoning Models (LRMs) have advanced beyond traditional Large Language Models, yet they pose heightened safety risks. |
| Approach: | They propose a first jailbreak attack targeting Large Reasoning Models . they exploit a Chaos Machine component to transform attack prompts with diverse one-to-one mappings based on the reasoning chain . |
| Outcome: | The proposed attack exploits the unique vulnerabilities of LRMs by integrating a Chaos Machine. success rates of the mousetrap attack are as high as 96%, 86% and 98% respectively. |
CachePrune: Teaching LLMs What Not to Follow via KV-Cache Editing (2026.acl-long)
Copied to clipboard
Rui Wang, Junda Wu, Yu Xia, Tong Yu, Ruiyi Zhang, Ryan A. Rossi, Subrata Mitra, Lina Yao, Julian McAuley
| Challenge: | Existing Large Language Models exhibit critical vulnerability to indirect prompt injection attacks, where instructions injected within in the prompt context can override the user's intent. |
| Approach: | They propose a neural pruning algorithm that prunes neurons associated with instruction-following during KV cache encoding of the prompt context. |
| Outcome: | The proposed approach significantly reduces the attack success rate while preserving the model's ability to follow user instructions. |
TensorOpera Router: A Multi-Model Router for Efficient LLM Inference (2024.emnlp-industry)
Copied to clipboard
Dimitris Stripelis, Zhaozhuo Xu, Zijian Hu, Alay Shah, Han Jin, Yuhang Yao, Jipeng Zhang, Tong Zhang, Salman Avestimehr, Chaoyang He
| Challenge: | Large Language Models (LLMs) have demonstrated remarkable performance across a diverse set of domain-specific tasks. |
| Approach: | They propose a non-monolithic LLM querying system that seamlessly integrates various LLM experts into a single query interface and dynamically routes incoming queries to the most high-performant expert based on query’s requirements. |
| Outcome: | The proposed model improves query efficiency by 40% and costs by 30% while maintaining or enhancing model performance by 10%. |
Generate-on-Graph: Treat LLM as both Agent and KG for Incomplete Knowledge Graph Question Answering (2024.emnlp-main)
Copied to clipboard
Yao Xu, Shizhu He, Jiabei Chen, Zihao Wang, Yangqiu Song, Hanghang Tong, Guang Liu, Jun Zhao, Kang Liu
| Challenge: | Existing methods to integrate LLMs with Knowledge Graphs (KGs) however, these methods are often incomplete to cover all the knowledge required to answer questions. |
| Approach: | They propose to integrate LLMs with Knowledge Graphs (KGs) to address insufficient knowledge and hallucination issues in Large Language Models. |
| Outcome: | The proposed method outperforms existing methods on two datasets. |
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. |
Mitigating Visual Knowledge Forgetting in MLLM Instruction-tuning via Modality-decoupled Gradient Descent (2025.findings-emnlp)
Copied to clipboard
Junda Wu, Yuxin Xiong, Xintong Li, Yu Xia, Ruoyu Wang, Yu Wang, Tong Yu, Sungchul Kim, Ryan A. Rossi, Lina Yao, Jingbo Shang, Julian McAuley
| Challenge: | Existing fine-tuning and continual learning methods compress visual representations and emphasize task alignment over visual retention. |
| Approach: | They propose a modality-decoupled gradient descent (MDGD) that regulates gradient updates to preserve effective rank of visual features and explicitly disentangles visual learning from task-specific alignment. |
| Outcome: | The proposed model reduces visual forgetting and improves visual retention . it disentangles visual learning from task-specific alignment and preserves effective rank . |
DIVKNOWQA: Assessing the Reasoning Ability of LLMs via Open-Domain Question Answering over Knowledge Base and Text (2024.findings-naacl)
Copied to clipboard
| Challenge: | Retrievalaugmented LLMs have been used to ground LLM in external knowledge . a gap exists in the current landscape regarding the effectiveness of grounding LLM on heterogeneous knowledge sources. |
| Approach: | They propose a model that uses symbolic language to generate symbolic queries . they use a dataset that is generated using predefined reasoning chains and human annotation . |
| Outcome: | The proposed model outperforms previous approaches by a significant margin in QA tasks over text. |
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. |
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. |
Personalized Federated Learning for Text Classification with Gradient-Free Prompt Tuning (2024.findings-naacl)
Copied to clipboard
Rui Wang, Tong Yu, Ruiyi Zhang, Sungchul Kim, Ryan Rossi, Handong Zhao, Junda Wu, Subrata Mitra, Lina Yao, Ricardo Henao
| Challenge: | Pretrained language models (PLMs) are used for personalized federated learning . communication costs are high with large PLMs, and local training is expensive . |
| Approach: | They propose a framework for federated learning with pretrained language models . they propose 'discrete local search' and compression mechanism for local training . |
| Outcome: | The proposed framework achieves superior performance compared with baselines. |
PRIME: A Process-Outcome Alignment Benchmark for Verifiable Reasoning in Mathematics and Engineering (2026.acl-long)
Copied to clipboard
Xiangfeng Wang, Hangyu Guo, Yanlin Lai, Mitt Huang, Liang Zhao, Chengyuan Yao, Yinmin Zhang, Qi Han, null Xiaoxiaoren, Chun Yuan, Tong Xu, Zheng Ge, Xiangyu Zhang, Daxin Jiang
| Challenge: | Current outcome-centric verification paradigms neglect potential errors in the derivation process. |
| Approach: | They propose a process-aware RLVR training paradigm utilizing verifiers selected via **PRIME**. |
| Outcome: | The proposed approach outperforms the baseline verification paradigm on AIME24, AIME25, and Beyond-AIME models. |
Cut the Deadwood Out: Backdoor Purification via Guided Module Substitution (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Model NLP models are often trained on datasets from untrusted platforms, posing significant risks of data poisoning attacks. |
| Approach: | They propose a retraining-free method that selectively replaces modules in the victim model based on a trade-off signal between utility and backdoor. |
| Outcome: | The proposed method outperforms even the strongest defense baseline against challenging attacks like LWS. |
From Individual Excellence to Collective Sustainability: Seeking Strategic Equilibrium in Proactive Multi-Agent Teams (2026.findings-acl)
Copied to clipboard
| Challenge: | a team of proactive agents suffer from a greedy optimization for immediate task accuracy . a new approach to improve team collaboration is based on the opportunity cost . |
| Approach: | They propose a game-theoretic proactive multi-agent reinforcement learning framework to solve this imbalance . they use a Positive-Unlabeled scorer to anchor intervention quality under sparse supervision . |
| Outcome: | The proposed framework maintains high performance while preventing experts from over-developing. |
How to Make LMs Strong Node Classifiers? (2026.findings-eacl)
Copied to clipboard
Zhe Xu, Kaveh Hassani, Si Zhang, Hanqing Zeng, Michihiro Yasunaga, Limei Wang, Dongqi Fu, Ning Yao, Bo Long, Hanghang Tong
| Challenge: | Language Models (LMs) are increasingly challenging the dominance of domain-specific models, such as Graph Neural Networks (GNNs) and Graph Transformers (GTs). |
| Approach: | They propose a novel approach that empowers off-the-shelf LMs to achieve performance comparable to state-of-the art (SOTA) GNNs on node classification tasks without requiring any architectural modifications. |
| Outcome: | The proposed approach outperforms existing GNNs on node classification tasks and is open-source upon publication. |
GLIMPSE: Do Large Vision-Language Models Truly Think With Videos or Just Glimpse at Them? (2025.emnlp-main)
Copied to clipboard
Yiyang Zhou, Linjie Li, Shi Qiu, Zhengyuan Yang, Yuyang Zhao, Siwei Han, Yangfan He, Kangqi Li, Haonian Ji, Zihao Zhao, Haibo Tong, Lijuan Wang, Huaxiu Yao
| Challenge: | Existing video benchmarks often resemble image-based questions with scans of only a few key frames, without deep temporal reasoning. |
| Approach: | They propose a video benchmark to assess whether large vision-language models can genuinely think with videos rather than perform superficial frame-level analysis. |
| Outcome: | The proposed benchmark consists of 3,269 videos and over 4,342 highly visual-centric questions across 11 categories, including Trajectory Analysis, Temporal Reasoning, and Forensics Detection. |
FANS: Formal Answer Selection for LLM Natural Language Math Reasoning Using Lean4 (2025.emnlp-main)
Copied to clipboard
| Challenge: | Existing frameworks that use Lean4 to enhance LLMs' NL reasoning abilities have been controversial in the field of math reasoning. |
| Approach: | They propose a framework that utilizes Lean4 to enhance LLMs’ NL math reasoning ability by generating a Lean 4 theorem statement and a proof-generating LLM. |
| Outcome: | The proposed framework improves LLMs' NL math reasoning ability by 2% across several math benchmarks and higher further based on reward models or in subfields such as algebra and number theory. |
ARC: Active and Reflection-driven Context Management for Long-Horizon Information Seeking Agents (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing approaches to managing context are based on raw accumulation or passive summarization, treating it as static artifact and allowing early errors or misplaced emphasis to persist. |
| Approach: | They propose a framework that treats context as a dynamic internal reasoning state during execution. |
| Outcome: | Experiments on long-horizon information-seeking benchmarks show that ARC outperforms passive context compression methods. |
Embedding-Informed Adaptive Retrieval-Augmented Generation of Large Language Models (2025.coling-main)
Copied to clipboard
| Challenge: | Retrieval-augmented large language models excel in various NLP tasks but are not always helpful when the knowledge required is absent in the model. |
| Approach: | They propose to determine whether the model is knowledgeable on a query via inspecting the (contextualized) pre-trained token embeddings of LLMs. |
| Outcome: | Experiments show that the proposed approach performs better than previous approaches on various benchmarks. |
CoMMIT: Coordinated Multimodal Instruction Tuning (2025.emnlp-main)
Copied to clipboard
Xintong Li, Junda Wu, Tong Yu, Rui Wang, Yu Wang, Xiang Chen, Jiuxiang Gu, Lina Yao, Julian McAuley, Jingbo Shang
| Challenge: | et al., 2024) show that multimodal instruction tuning is more effective than baselines. |
| Approach: | They propose a multimodal balance coefficient that enables quantitative measurement of the balance of learning . they propose auxiliary regularization on the gradient to promote updating with larger step sizes . |
| Outcome: | The proposed method is more effective than baselines in MLLM instruction tuning. |
GUI Agents: A Survey (2025.findings-acl)
Copied to clipboard
Dang Nguyen, Jian Chen, Yu Wang, Gang Wu, Namyong Park, Zhengmian Hu, Hanjia Lyu, Junda Wu, Ryan Aponte, Yu Xia, Xintong Li, Jing Shi, Hongjie Chen, Viet Dac Lai, Zhouhang Xie, Sungchul Kim, Ruiyi Zhang, Tong Yu, Mehrab Tanjim, Nesreen K. Ahmed, Puneet Mathur, Seunghyun Yoon, Lina Yao, Branislav Kveton, Jihyung Kil, Thien Huu Nguyen, Trung Bui, Tianyi Zhou, Ryan A. Rossi, Franck Dernoncourt
| Challenge: | Large Foundation Models (LFMs) have transformed the landscape of AI research and day-to-day life. |
| Approach: | They propose a framework that delineates GUI agents' perception, reasoning, planning, and acting capabilities. |
| Outcome: | The proposed framework delineates their perception, reasoning, planning, and acting capabilities. |
VecInfer: Efficient LLM Inference with Low-Bit KV Cache via Outlier-Suppressed Vector Quantization (2026.acl-long)
Copied to clipboard
| Challenge: | Existing quantization methods for large language models suffer performance degradation at ultra-low bit-widths due to key cache outliers. |
| Approach: | They propose a vector quantization method that suppresses outliers in the key cache and reduces memory access overhead. |
| Outcome: | The proposed method outperforms baseline quantization methods across long-context understanding and mathematical reasoning tasks while minimizing memory access overhead. |
SAND: Boosting LLM Agents with Self-Taught Action Deliberation (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large Language Model (LLM) agents finetuned with supervised finetuning may over-commit towards seemingly plausible but suboptimal actions due to limited action space exploration. |
| Approach: | They propose a self-taught actioN deliberation framework that allows LLM agents to explicitly deliberate over candidate actions before committing to one. |
| Outcome: | The proposed framework outperforms state-of-the-art methods on two representative interactive agent tasks and achieves an average 20% improvement over initial finetuning. |
iPET: An Interactive Emotional Companion Dialogue System with LLM-Powered Virtual Pet World Simulation (2025.acl-demo)
Copied to clipboard
Zheyong Xie, Shaosheng Cao, Zuozhu Liu, Zheyu Ye, Zihan Niu, Chonggang Lu, Tong Xu, Enhong Chen, Zhe Xu, Yao Hu, Wei Lu
| Challenge: | Existing approaches to role-playing emotional companion products lack sustained personalization and contextual adaptability, limiting their effectiveness in real-world settings. |
| Approach: | They propose a virtual pet agent that can enhance user engagement through rich, dynamic pet behaviors and interactions tailored to individual preferences. |
| Outcome: | The proposed system has been deployed in a real-world, non-commercial product for 200 days and has demonstrated its effectiveness in practical applications. |