Papers by Furong Huang
Uncertainty-Aware Answer Selection for Improved Reasoning in Multi-LLM Systems (2025.findings-emnlp)
Copied to clipboard
Aakriti Agrawal, Rohith Aralikatti, Anirudh Satheesh, Souradip Chakraborty, Amrit Singh Bedi, Furong Huang
| Challenge: | Existing approaches to selecting reliable responses from multiple LLMs often depend on external verifiers, human evaluators, or self-consistency techniques. |
| Approach: | They propose a calibrated log-likelihood-based selection framework to improve multi-LLM performance. |
| Outcome: | The proposed method outperforms majority voting and exceeds self-consistency performance when using a large number of model calls. |
MergeME: Model Merging Techniques for Homogeneous and Heterogeneous MoEs (2025.naacl-long)
Copied to clipboard
Yuhang Zhou, Giannis Karamanolakis, Victor Soto, Anna Rumshisky, Mayank Kulkarni, Furong Huang, Wei Ai, Jianhua Lu
| Challenge: | State-of-the-art methods for merging expert models with different architectures do not address parameter interference and require extensive fine-tuning to restore performance. |
| Approach: | They propose a method for merging experts with different architectures into a unified Mixture-of-Experts model with a goal of enhancing performance in each domain while retaining effectiveness on general tasks. |
| Outcome: | Experiments across multiple domains show that the proposed methods reduce fine-tuning costs and improve performance over state-of-the-art methods. |
Teach a Reward Model to Correct Itself: Reward Guided Adversarial Failure Discovery for Robust Reward Modeling (2026.acl-long)
Copied to clipboard
| Challenge: | Existing failure discovery methods rely on prior knowledge of preference attributes . Existing methods do not scale to new models or data. |
| Approach: | They propose a preference distribution agnostic procedure that uses the reward model itself to guide controlled decoding toward mis specified responses while preserving the underlying preference class. |
| Outcome: | The proposed procedure improves robustness without degrading reward quality across models. |
Explore Spurious Correlations at the Concept Level in Language Models for Text Classification (2024.acl-long)
Copied to clipboard
| Challenge: | Language models have demonstrated remarkable performance in numerous NLP tasks, employing both fine-tuning and in-context learning (ICL) methods. |
| Approach: | They propose a method to assess concept bias in models during fine-tuning and in-context learning using ChatGPT. |
| Outcome: | The proposed method outperforms token removal approaches and is validated through extensive testing. |
AutoHallusion: Automatic Generation of Hallucination Benchmarks for Vision-Language Models (2024.findings-emnlp)
Copied to clipboard
Xiyang Wu, Tianrui Guan, Dianqi Li, Shuaiyi Huang, Xiaoyu Liu, Xijun Wang, Ruiqi Xian, Abhinav Shrivastava, Furong Huang, Jordan Boyd-Graber, Tianyi Zhou, Dinesh Manocha
| Challenge: | Large vision-language models are prone to hallucinations, where contextual cues in an image can trigger the language module to produce overconfident and incorrect reasoning about abnormal or hypothetical objects. |
| Approach: | They propose to automate the generation of hallucination-related questions using images . they propose to use three image manipulation strategies to induce hallucinosity . |
| Outcome: | The proposed approach reduces human bias in crafting such examples and improves accuracy. |
Jailbreaks as Inference-Time Alignment: A Framework for Understanding Safety Failures in LLMs (2026.eacl-long)
Copied to clipboard
| Challenge: | Large language models are safety-aligned to prevent harmful response generation . prior work on jailbreak effectiveness has focused on analyzing success rate of jailbreaks . |
| Approach: | They propose to frame jailbreaks as inference-time alignment and draw suboptimal bounds . they also propose a Safety-Net to measure how vulnerable an LLM is to jailbreak attacks . |
| Outcome: | a new framework allows researchers to show how vulnerable an LLM is to jailbreaks . a Safety-Net measures how vulnerable the model is to attacks, the authors say . |
Large Language Models and Causal Inference in Collaboration: A Comprehensive Survey (2025.findings-naacl)
Copied to clipboard
Xiaoyu Liu, Paiheng Xu, Junda Wu, Jiaxin Yuan, Yifan Yang, Yuhang Zhou, Fuxiao Liu, Tianrui Guan, Haoliang Wang, Tong Yu, Julian McAuley, Wei Ai, Furong Huang
| Challenge: | Large Language Models (LLMs) have shown great potential to enhance Natural Language Processing (NLP) models in areas such as predictive accuracy, fairness, robustness, and explainability. |
| Approach: | They evaluate or improve generative Large Language Models from a causal perspective in areas such as reasoning capacity, fairness and safety issues, explainability, and handling multimodality. |
| Outcome: | The proposed models can be used to perform causal relationship discovery and causal effect estimation tasks. |
PoisonedParrot: Subtle Data Poisoning Attacks to Elicit Copyright-Infringing Content from Large Language Models (2025.naacl-long)
Copied to clipboard
Michael-Andrei Panaitescu-Liess, Pankayaraj Pathmanathan, Yigitcan Kaya, Zora Che, Bang An, Sicheng Zhu, Aakriti Agrawal, Furong Huang
| Challenge: | PoisonedParrot is the first stealthy data poisoning attack that induces an LLM to generate copyrighted content even when the model has not been directly trained on the copyright material. |
| Approach: | They propose a stealthy data poisoning attack that induces an LLM to generate copyrighted content even when it has not been directly trained on the copyright material. |
| Outcome: | The proposed model induces an LLM to generate copyrighted content with no discernible side effects and is surprisingly effective at priming the model to generate content with little side effects. |
DISCO Balances the Scales: Adaptive Domain- and Difficulty-Aware Reinforcement Learning on Imbalanced Data (2025.findings-emnlp)
Copied to clipboard
Yuhang Zhou, Jing Zhu, Shengyi Qian, Zhuokai Zhao, Xiyao Wang, Xiaoyu Liu, Ming Li, Paiheng Xu, Wei Ai, Furong Huang
| Challenge: | Large Language Models (LLMs) are increasingly aligned with human preferences through Reinforcement Learning from Human Feedback (RLHF). |
| Approach: | a new study proposes a domain-informed self-consistency policy optimization extension to GRPO that addresses inter-group imbalance. |
| Outcome: | a new extension of GRPO addresses inter-group imbalance with two key innovations . the proposed method outperforms existing GR PO variants by 5% on Qwen3 models . |
World Models with Hints of Large Language Models for Goal Achieving (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing methods address this by adding intrinsic rewards, but they fail to provide meaningful guidance in long-horizon decision-making tasks with large state and action spaces lacking purposeful exploration. |
| Approach: | They propose a multi-modal model-based RL approach that integrates the proposed hinting subgoals into the model rollouts to encourage goal discovery and reaching in challenging tasks. |
| Outcome: | The proposed model outperforms existing methods in challenging, sparse-reward environments such as HomeGrid, Crafter, and Minecraft by 41.8%, 21.1%, and 9.9%. |
Mementos: A Comprehensive Benchmark for Multimodal Large Language Model Reasoning over Image Sequences (2024.acl-long)
Copied to clipboard
Xiyao Wang, Yuhang Zhou, Xiaoyu Liu, Hongjin Lu, Yuancheng Xu, Feihong He, Jaehong Yoon, Taixi Lu, Fuxiao Liu, Gedas Bertasius, Mohit Bansal, Huaxiu Yao, Furong Huang
| Challenge: | Multimodal Large Language Models (MLLMs) have demonstrated proficiency in handling a variety of visual-language tasks, but their ability to extrapolate from image sequences has been less investigated. |
| Approach: | They propose a new benchmark to assess MLLMs’ sequential image reasoning abilities. |
| Outcome: | The proposed benchmark features 4,761 diverse image sequences with varying lengths. |
Towards Mitigating Hallucinations in Large Vision-Language Models by Refining Textual Embeddings (2026.findings-acl)
Copied to clipboard
Aakriti Agrawal, Gouthaman KV, Rohith Aralikatti, Gauri Jagatap, Jiaxin Yuan, Sarvesh Baskar, Vijay Kamarshi, Andrea Fanelli, Furong Huang
| Challenge: | Hallucinations in Large Vision-Language Models (LVLMs) are a persistent challenge, stemming from inadequate integration of visual information during multimodal reasoning. |
| Approach: | They propose a visual feature incorporation method that encourages the model to learn visually-informed textual embeddings distinct from those of the base LLM and promotes a more balanced attention distribution. |
| Outcome: | The proposed method significantly reduces hallucinations and fosters more balanced multimodal reasoning. |
EnsemW2S: Enhancing Weak-to-Strong Generalization with Large Language Model Ensembles (2026.findings-acl)
Copied to clipboard
Aakriti Agrawal, Mucong Ding, Chenghao Deng, Zora Che, Arjun Rajaram, Anirudh Satheesh, Bang An, C. Bayan Bruss, John Langford, Furong Huang
| Challenge: | Large Language Models (LLMs) are rapidly approaching and potentially exceeding human-level performance . a novel method aims to improve weak experts' generalization abilities by training them on limited human- level data . |
| Approach: | They propose a method that iteratively combines multiple weak experts to improve their generalization performance by training on limited human-level data. |
| Outcome: | The proposed method improves weak experts' generalization abilities by iterating on weak models and stronger student models. |
Multi-Stage Balanced Distillation: Addressing Long-Tail Challenges in Sequence-Level Knowledge Distillation (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Knowledge distillation (KD) is a promising solution for large language models, but their deployment remains computationally expensive. |
| Approach: | They propose a framework which iteratively balances training data within a fixed computational budget and enables the transfer of knowledge from expensive teacher LLMs to smaller student models. |
| Outcome: | The proposed framework achieves state-of-the-art performance across diverse long-tailed datasets, enhancing both the efficiency and efficacy of the distilled models. |
Enhancing Visual-Language Modality Alignment in Large Vision Language Models via Self-Improvement (2025.findings-naacl)
Copied to clipboard
Xiyao Wang, Jiuhai Chen, Zhaoyang Wang, Yuhang Zhou, Yiyang Zhou, Huaxiu Yao, Tianyi Zhou, Tom Goldstein, Parminder Bhatia, Taha Kass-Hout, Furong Huang, Cao Xiao
| Challenge: | Existing methods for visual and language alignment depend on external models or data, leading to uncontrollable and unstable results. |
| Approach: | They propose a framework that enhances visual and language alignment without external dependencies by incorporating an in-context self-critic mechanism that constructs preference pairs for tuning. |
| Outcome: | The proposed framework outperforms existing methods and improves performance on 14 hallucination and comprehensive benchmarks. |