Papers by Jingyu Zhang
PunMemeCN: A Benchmark to Explore Vision-Language Models’ Understanding of Chinese Pun Memes (2025.emnlp-main)
Copied to clipboard
| Challenge: | Pun memes combine wordplay with visual elements to create humor, irony, or other rhetorical effects. |
| Approach: | They propose a benchmark to assess Chinese pun memes' processing capabilities across three progressive tasks: pun meme detection, sentiment analysis, and chat-driven meme response. |
| Outcome: | The proposed model can detect pun memes, analyze sentiments, and respond to chats, while ignoring homophone wordplay. |
Rethinking Smoothness for Fast and Adaptable Entity Alignment Decoding (2025.findings-naacl)
Copied to clipboard
| Challenge: | Existing methods for integrating knowledge graphs rely on entity and relation embeddings . Fig. 1 shows how to decode knowledge graph in under 6 seconds . |
| Approach: | They propose a framework that only utilizes entity embeddings to decode knowledge graphs. |
| Outcome: | The proposed framework reconstructs KG representation by maximizing smoothness of entity embeddings. |
VET: Verifiable Execution Tracing for Reliable Text-to-SQL Generation (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing methods for text-to-SQL generation are prone to hallucinations and grounding . authors present a novel reasoning paradigm that transforms text- to-Sql from unverifiable textual rationales into step-wise executable semantics. |
| Approach: | They propose a reasoning paradigm that transforms text-to-SQL from unverifiable textual rationales into step-wise executable semantics. |
| Outcome: | The proposed reasoning paradigm transforms text-to-SQL from unverifiable textual rationales into step-wise executable semantics. |
VALU: A Benchmark for Video Anomaly Temporal Localization and Understanding at Multiple Semantic Levels (2026.acl-long)
Copied to clipboard
Yixiao He, Menghao Zhang, Haifeng Sun, Jing Wang, Kangheng Lin, Jinghan Wang, Chenye Xu, Pengfei Ren, Qi Qi, Jingyu Wang
| Challenge: | Recent advances in Video Large Language Models (Video-LLMs) enhance the ability of VAU models to describe and interpret anomalies. |
| Approach: | They propose a benchmark that explicitly defines anomalies across five semantic levels and provides detailed temporal boundaries and detailed textual descriptions for each. |
| Outcome: | The proposed benchmark defines anomalies across five semantic levels and provides detailed descriptions for each. |
ClusterAttn: KV Cache Compression under Intrinsic Attention Clustering (2025.acl-long)
Copied to clipboard
| Challenge: | Existing methods for sparse attention apply the same pattern across different attention heads and inputs, but fail to capture the intrinsic attention clustering in large language models. |
| Approach: | They propose a training-free sparse attention method that provides an efficient prompt cache compression scheme under intrinsic attention clustering for efficient LLM inference. |
| Outcome: | The proposed method reduces memory usage by 10%–65% and increases throughput by 2.6–4.8 times with no accuracy loss. |
Beyond Reasoning Gains: Mitigating General-Capability Forgetting in Large Reasoning Models (2026.findings-acl)
Copied to clipboard
Hoang Phan, Xianjun Yang, Yuanshun Yao, Jingyu Zhang, Shengjie Bi, Xiaocheng Tang, Madian Khabsa, Lijuan Liu, Deren Lei
| Challenge: | Reinforcement learning with verifiable rewards (RLVR) has delivered impressive gains in mathematical and multimodal reasoning . however, the recipe introduces a significant risk of capability regression, where models forget foundational skills after prolonged training without employing regularization strategies. |
| Approach: | They propose a replay strategy with dynamic objective reweighting for general knowledge preservation using short-horizon signals of convergence and instability. |
| Outcome: | The proposed method preserves general capabilities and improves reasoning . it can be applied to existing RLVR pipelines without training additional models or tuning . |
Core: Robust Factual Precision with Informative Sub-Claim Identification (2025.findings-acl)
Copied to clipboard
Zhengping Jiang, Jingyu Zhang, Nathaniel Weir, Seth Ebner, Miriam Wanner, Kate Sanders, Daniel Khashabi, Anqi Liu, Benjamin Van Durme
| Challenge: | Using the Decompose-Then-Verify framework, such as FActScore, can be manipulated by adding obvious or repetitive subclaims to artificially inflate scores. |
| Approach: | They propose a decomposition-based tool called Core to filter subclaims based on their uniqueness and informativeness. |
| Outcome: | The proposed evaluation framework supports easy and modular use of Core and various decomposition strategies. |
Jailbreak Distillation: Renewable Safety Benchmarking (2025.findings-emnlp)
Copied to clipboard
Jingyu Zhang, Ahmed Elgohary, Xiawei Wang, A S M Iftekhar, Ahmed Magooda, Benjamin Van Durme, Daniel Khashabi, Kyle Jackson
| Challenge: | a new framework for safety benchmarking is being developed for large language models . the framework "distills" jailbreak attacks into high-quality and easily-updatable safety benchmarks . |
| Approach: | They propose a framework that "distills" jailbreak attacks into high-quality safety benchmarks. |
| Outcome: | The proposed framework "distills" jailbreak attacks into high-quality safety benchmarks . it requires minimal human effort to rerun the pipeline and produce updated benchmarks. |
SSS: Editing Factual Knowledge in Language Models towards Semantic Sparse Space (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to modify LMs suffer from sub-optimal locality, where irrelevant neighborhood examples can be adversely influenced. |
| Approach: | They propose to use a model editing method to modify specific examples in LMs to improve locality and reasoning capability by directing the hidden state of edit example towards spaces where semantics are sparse. |
| Outcome: | The proposed method improves locality and reasoning capability on two datasets. |
JARVIS or Ultron? A Survey on the Safety and Security Threats of Computer-Using Agents (2026.acl-long)
Copied to clipboard
Ada Chen, Yongjiang Wu, Junyuan Zhang, Jingyu Xiao, Shu Yang, Jen-tse Huang, Kun Wang, Wenxuan Wang, Shuai Wang
| Challenge: | Recent advances in computer-using agents have created new safety and security risks . despite the impressive capabilities of CUAs, there are still significant security risks. |
| Approach: | They propose a systematization of knowledge on the safety and security threats of Computer-Using Agents. |
| Outcome: | The proposed framework provides a framework for assessing the safety and security risks of computer-using agents. |
Geo-Seq2seq: Twitter User Geolocation on Noisy Data through Sequence to Sequence Learning (2023.findings-acl)
Copied to clipboard
| Challenge: | a new method for Twitter user geolocation rewrites noisy, multilingual location strings into structured English location names. |
| Approach: | They propose a sequence-to-sequence (seq2sequ) model that rewrites noisy location strings into structured English location names. |
| Outcome: | The proposed model can generalize well to unseen temporal data, but performance does vary by language and country. |
TurkingBench: A Challenge Benchmark for Web Agents (2025.naacl-long)
Copied to clipboard
Kevin Xu, Yeganeh Kordi, Tanay Nayak, Adi Asija, Yizhong Wang, Kate Sanders, Adam Byerly, Jingyu Zhang, Benjamin Van Durme, Daniel Khashabi
| Challenge: | TurkingBench is a benchmark consisting of tasks presented as web pages with textual instructions and multi-modal contexts. |
| Approach: | They propose to use HTML pages to perform various annotation tasks on crowdsourcing platforms. |
| Outcome: | The proposed model outperforms other models on the TurkingBench benchmark. |
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. |
k-SemStamp: A Clustering-Based Semantic Watermark for Detection of Machine-Generated Text (2024.findings-acl)
Copied to clipboard
| Challenge: | Recent watermarked generation algorithms inject detectable signatures during language generation to facilitate post-hoc detection. |
| Approach: | They propose a watermark which assigns signatures to each watermarked sentence according to locality-sensitive hashing (LSH) they propose k-SemStamp, which uses kmeans clustering to partition the semantic space with awareness of inherent semantic structure. |
| Outcome: | The proposed watermark improves its robustness and sampling efficiency while preserving the generation quality, making it more effective for machine-generated text detection. |
Verifiable by Design: Aligning Language Models to Quote from Pre-Training Data (2025.naacl-long)
Copied to clipboard
| Challenge: | Recent efforts to verify text accuracy provide no guarantees on their correctness . a new method to improve LLMs' verifiability is to use quotes to ground models . |
| Approach: | They propose a method that allows models to quote verbatim statements from trusted sources . they leverage a fast membership inference function to verify text against trusted corpora . |
| Outcome: | The proposed method significantly increases verbatim quotes from high-quality documents by up to 130% relative to base models while maintaining response quality. |
Versatile Framework for Song Generation with Prompt-based Control (2025.findings-emnlp)
Copied to clipboard
Yu Zhang, Wenxiang Guo, Changhao Pan, Zhiyuan Zhu, Ruiqi Li, Jingyu Lu, Rongjie Huang, Ruiyuan Zhang, Zhiqing Hong, Ziyue Jiang, Zhou Zhao
| Challenge: | Existing methods for song generation fail to generate vocals with prompt-based control and proper alignment. |
| Approach: | VersBand is a multi-task song generation framework for synthesizing high-quality songs with prompt-based control. |
| Outcome: | Experimental results show that VersBand performs better than baseline models across multiple song generation tasks. |
NaviMaster: Learning a Unified Policy for GUI and Embodied Navigation Tasks (2026.acl-long)
Copied to clipboard
| Challenge: | Recent advances in Graphical User Interface (GUI) and embodied navigation have driven progress, yet these domains have largely evolved in isolation, with disparate datasets and training paradigms. |
| Approach: | They propose a visual-target trajectory collection pipeline that generates trajectories for GUI and embodied tasks using a single formulation. |
| Outcome: | The proposed agent outperforms state-of-the-art agents in GUI navigation, spatial affordance prediction, and embodied navigation. |
RATIONALYST: Pre-training Process-Supervision for Improving Reasoning (2025.acl-long)
Copied to clipboard
Dongwei Jiang, Guoxuan Wang, Yining Lu, Andrew Wang, Jingyu Zhang, Chuyu Liu, Benjamin Van Durme, Daniel Khashabi
| Challenge: | RATIONALYST is a model for process-supervision of reasoning based on pretraining on rationale annotations extracted from unlabeled data. |
| Approach: | They propose a model for process-supervision of reasoning based on pre-training on rationale annotations extracted from unlabeled data. |
| Outcome: | RATIONALYST improves reasoning accuracy by 3.9% on representative reasoning benchmarks. |
Effective Long-Context Scaling of Foundation Models (2024.naacl-long)
Copied to clipboard
Wenhan Xiong, Jingyu Liu, Igor Molybog, Hejia Zhang, Prajjwal Bhargava, Rui Hou, Louis Martin, Rashi Rungta, Karthik Abinav Sankararaman, Barlas Oguz, Madian Khabsa, Han Fang, Yashar Mehdad, Sharan Narang, Kshitiz Malik, Angela Fan, Shruti Bhosale, Sergey Edunov, Mike Lewis, Sinong Wang, Hao Ma
| Challenge: | Large language models (LLMs) are rapidly deployed and continue to evolve through scaling. |
| Approach: | They propose a method to train strong long-context LLMs that are capable of utilizing massive context windows of up to 32,000 tokens. |
| Outcome: | The proposed model can surpass gpt-3.5-turbo-16k's overall performance on long-context benchmarks with a cost-effective instruction tuning procedure that is free of expensive annotations. |
On the Blind Spots of Model-Based Evaluation Metrics for Text Generation (2023.acl-long)
Copied to clipboard
| Challenge: | Existing methods for text generation evaluation metrics are lacking in robustness analysis. |
| Approach: | They propose to use stress tests to test for errors in text generation evaluation metrics . they find that BERTScore is confused by truncation errors in summarization . |
| Outcome: | The proposed stress tests show that they are insensitive to errors in open-ended generation, translation, and summarization. |
On the Zero-Shot Generalization of Machine-Generated Text Detectors (2023.findings-emnlp)
Copied to clipboard
| Challenge: | rampant proliferation of large language models generates text indistinguishable from human-written language. |
| Approach: | They train neural detectors on outputs of a new generator and test their performance on held-out generators. |
| Outcome: | The proposed detectors can be built on training data from medium-sized models. |
PCFG-Based Natural Language Interface Improves Generalization for Controlled Text Generation (2023.starsem-1)
Copied to clipboard
| Challenge: | Existing work on controlled text generation (CTG) assumes a control interface of categorical attributes. |
| Approach: | They propose a natural language interface to embed control attributes into natural language commands and propose variants of existing CTG models that take commands as input. |
| Outcome: | The proposed model can generalize to unseen attributes and unsealed attribute combinations. |
Certified Mitigation of Worst-Case LLM Copyright Infringement (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models are trained on vast datasets that include copyrighted material or content with usage restrictions. |
| Approach: | They propose a "copyright takedown" method that provides certified copyright take down . they use a combination of inference-time and rewriting techniques to transform potentially infringing segments . |
| Outcome: | The proposed method reduces infringement risk, preserves utility, and accommodates different levels of enforcement stringency with adaptive abstention. |
Stepwise Reasoning Disruption Attack of LLMs (2025.acl-long)
Copied to clipboard
Jingyu Peng, Maolin Wang, Xiangyu Zhao, Kai Zhang, Wanyu Wang, Pengyue Jia, Qidong Liu, Ruocheng Guo, Qi Liu
| Challenge: | Existing attacks on LLM reasoning are constrained by specific settings or lack of imperceptibility, limiting their feasibility and generalizability. |
| Approach: | They propose a stepwise rEasoning error disruption attack that subtly injects errors into prior reasoning steps to mislead the model into producing incorrect subsequent reasoning and final answers. |
| Outcome: | The proposed attack is compatible with zero-shot and few-shot settings, maintains the natural reasoning flow, and ensures covert execution without modification of the instruction. |
JailMeter: An Evidence-Based Evaluation Framework for Jailbreak Attacks on Large Language Models (2026.findings-acl)
Copied to clipboard
Qingjia Huang, Jingyu Zhang, Jianguo Wu, Yakai Li, Weijuan Zhang, Yankai Rong, Junyi Yao, Shengzhi Zhang, Xiaoqi Jia
| Challenge: | Currently, evaluation criteria and methods used for jailbreak effectiveness are inconsistent. |
| Approach: | They propose a framework to measure jailbreak effectiveness using a model that filters out jailbreak noise while preserving the original malicious question. |
| Outcome: | The proposed framework outperforms existing evaluation methods on a challenging benchmark containing 330 human-labeled, non-rejected jailbreak instances. |
The Language Barrier: Dissecting Safety Challenges of LLMs in Multilingual Contexts (2024.findings-acl)
Copied to clipboard
Lingfeng Shen, Weiting Tan, Sihao Chen, Yunmo Chen, Jingyu Zhang, Haoran Xu, Boyuan Zheng, Philipp Koehn, Daniel Khashabi
| Challenge: | Recent studies show that malicious prompt instructions could solicit objectionable content from LLMs. |
| Approach: | They compare how state-of-the-art LLMs respond to malicious prompts in different languages . they find that LLM's generate unsafe responses more often when a prompt is written in a lower-resource language . |
| Outcome: | The proposed model can generate unsafe responses more often when a malicious prompt is written in a lower-resource language, and less irrelevant responses when written in lower-source languages. |
SemStamp: A Semantic Watermark with Paraphrastic Robustness for Text Generation (2024.naacl-long)
Copied to clipboard
Abe Hou, Jingyu Zhang, Tianxing He, Yichen Wang, Yung-Sung Chuang, Hongwei Wang, Lingfeng Shen, Benjamin Van Durme, Daniel Khashabi, Yulia Tsvetkov
| Challenge: | Existing watermarked generation algorithms employ token-level designs and are vulnerable to paraphrase attacks. |
| Approach: | They propose a sentence-level watermarking algorithm that uses locality-sensitive hashing to partition the semantic space of sentences. |
| Outcome: | The proposed algorithm is more robust than the existing state-of-the-art method on paraphrasers and domains, while posing only minor degradations to SemStamp. |
Logic Jailbreak: Efficiently Unlocking LLM Safety Restrictions Through Formal Logical Expression (2026.findings-acl)
Copied to clipboard
Jingyu Peng, Maolin Wang, Nan Wang, Jiatong Li, Yuchen Li, Yuyang Ye, Wanyu Wang, Pengyue Jia, Kai Zhang, Xiangyu Zhao
| Challenge: | Despite advances in aligning LLMs with human values, current safety mechanisms remain vulnerable to jailbreak attacks. |
| Approach: | They propose a black-box jailbreak method that uses logical expression translation to bypass LLM safety mechanisms. |
| Outcome: | The proposed method exploits the distributional gap between alignment data and logic-expressed inputs while preserving the underlying semantic intent and readability while evading safety constraints. |
Example Quality Matters: Multi-Aspects Example Augmentation for Private Library Programming (2026.acl-long)
Copied to clipboard
Yuhao Li, Haifeng Sun, Xuesong Zhang, Shu Yao, Haoyu Zheng, Yvchuan Wang, Huazheng Wang, Zirui Zhuang, Qi Qi, Jianxin Liao, Jingyu Wang
| Challenge: | Existing approaches to code generation fail to consider the quality of retrieved examples. |
| Approach: | They propose a retrieval-augmented generation method that combines existing API examples to improve complexity and readability. |
| Outcome: | The proposed method achieves up to 22% accuracy improvement over baseline methods. |