Papers by Geng Zhang

SGPVT: Self-Generated Proximal Visual Tokens for Mitigating Proximal Collateral Damage in MLLM Unlearning (2026.acl-long)

Challenge:	Existing models with reasoning capabilities suffer from a severe length collapse in open-ended writing .
Approach:	They propose a framework that embeds a dynamic plan-write-reflect cycle into the generation process and train a model with interleaved reasoning traces.
Outcome:	The proposed framework achieves state-of-the-art performance on long-form benchmarks compared to other models on the same dataset.

Copied to clipboard

Jiaqi Li, Zhijing Zhang, Jiahui Geng, Sheng Bi, Chuanyi Zhang, Fan Liu, Guilin Qi

DisCal: Distribution-Aware Calibration for Mathematical Reasoning Under Character-Level Noisy Inputs (2026.acl-long)

Challenge:	Existing approaches focus on general utility metrics, overlooking the preservation of semantically related concepts.
Approach:	They propose a method that introduces self-generated proximal visual tokens to prevent forgetting vulnerability.
Outcome:	The proposed framework outperforms existing methods in preserving semantically related concepts while achieving effective target unlearning.

Copied to clipboard

Bo Zhang, Jiawei Zhang, Cong Gao, Bingxu Han, Minghao Hu, Jun Zhang, Yunbo Cao, Zhunchen Luo, Wen Yao, Guotong Geng, Zhong Wang

ReasoningGuard: Safeguarding Large Reasoning Models with Inference-time Safety Aha Moments (2026.acl-long)

Challenge:	Existing methods for calibration of large reasoning models (LRMs) focus on clean inputs, leaving noise unexplored.
Approach:	They propose a confidence calibration framework for character-level noisy inputs that extracts uncertainty signals from both the empirical answer distribution and the model’s predictive distribution and integrates them via a learned calibrator.
Outcome:	Experiments on multiple mathematical reasoning benchmarks show that DisCal outperforms existing calibration methods under noisy inputs, reducing expected calibration error (ECE) by up to 39.21% and improving Area Under the Receiver Operating Characteristic Curve (AUROC) by 31.44%.

Copied to clipboard

Yuquan Wang, Mi Zhang, Yining Wang, Geng Hong, Mi Wen, Xiaoyu You, Min Yang

Always the Best Fit: Adaptive Domain Gap Filling from Causal Perspective for Few-Shot Relation Extraction (2023.findings-emnlp)

Challenge:	Existing defenses for Large Reasoning Models (LRMs) depend on costly fine-tuning and additional expert knowledge, which limits their scalability.
Approach:	They propose an inference-time safeguard for Large Reasoning Models that injects safety aha moments into the reasoning process to guide the model towards harmless yet helpful reasoning.
Outcome:	The proposed safeguard outperforms nine existing safeguards while avoiding common exaggerated safety issues.

Copied to clipboard

Ge Bai, Chenji Lu, Jiaxiang Geng, Shilong Li, Yidong Shi, Xiyan Liu, Ying Liu, Zhang Zhang, Ruifang Liu

Improving Personalized Explanation Generation through Visualization (2022.acl-long)

Challenge:	Existing approaches to cross-domain relation extraction have been limited by domains . data bias between domains can be difficult to fill, especially in few-shot scenarios .
Approach:	They propose a framework to bridge the semantic gap caused by data bias between domains . they use syntactic structure, label distribution, and entities to calculate causal effects .
Outcome:	The proposed framework fills the domain gap and yields better results on the few-shot task.

Copied to clipboard

Shijie Geng, Zuohui Fu, Yingqiang Ge, Lei Li, Gerard de Melo, Yongfeng Zhang

Explainability & Interpretability in NLP

Challenge:	Existing explainable recommendation models generate repetitive sentences for different items or empty sentences with insufficient details.
Approach:	They propose a visual-enhanced approach to generate rating scores and text explanations using visualization generation and text–image matching discrimination.
Outcome:	The proposed approach improves both the text quality and the diversity and explainability of the generated explanations.

Data-Efficient Selection via Grammatical Complexity in Continual Pre-training of Domain-Specific LLMs (2025.emnlp-main)

Copied to clipboard

Yizhou Ying, Geng Zhang, Cui Danxin, Chengyu Du, Guanglei Yue, Sihang Jiang, Jiaqing Liang, Yifei Fu, Hailin Hu, Yanghua Xiao

AlphaOne: Reasoning Models Thinking Slow and Fast at Test Time (2025.emnlp-main)

Challenge:	Existing data selection strategies for continual pre-training of large language models often rely on scarce labeled data or computationally expensive LLMs.
Approach:	They propose an annotation-independent data selection framework for CPT that evaluates grammatical complexity using lexical diversity and syntactic complexity.
Outcome:	The proposed framework outperforms baselines on a financial dataset and surpasses full-data training by 1.7% using only 20% of the data.

Copied to clipboard

Junyu Zhang, Runpei Dong, Han Wang, Xuying Ning, Haoran Geng, Peihao Li, Xialin He, Yutong Bai, Jitendra Malik, Saurabh Gupta, Huan Zhang

SHARP: Self-adaptive Harmful Category-aware Prompt Generation for Black-box Jailbreaking (2026.acl-long)

Challenge:	Existing monotonic scaling methods for large reasoning models are not reliable.
Approach:	They propose a universal framework for modulating reasoning progress in large reasoning models at test time.
Outcome:	The proposed framework unifies and generalizes existing monotonic scaling methods and enables flexible and dense slow-to-fast reasoning modulation.

Copied to clipboard

Yingjie Xue, Xingyou Xia, Jun Zhang, Yunbo Cao, Dengpan Ye, Guotong Geng, Fei Li

Marco-Bench-MIF: On Multilingual Instruction-Following Capability of Large Language (2025.acl-long)

Challenge:	Existing methods for jailbreak ignore the semantic differences between categories of harmful questions, leading to inconsistent success rates and reduced overall attack effectiveness.
Approach:	They propose a category-aware jailbreak framework that incorporates the semantic category of harmful questions into prompt generation.
Outcome:	The proposed framework improves attack success rates and category alignment and achieves better cross-category robustness compared to the state-of-the-art (SOTA) baselines.

Copied to clipboard

Bo Zeng, Chenyang Lyu, Sinuo Liu, Mingyan Zeng, Minghao Wu, Xuanfan Ni, Tianqi Shi, Yu Zhao, Yefeng Liu, Chenyu Zhu, Ruizhe Li, Jiahui Geng, Qing Li, Yu Tong, Longyue Wang, Weihua Luo, Kaifu Zhang

Lost in the Source Language: How Large Language Models Evaluate the Quality of Machine Translation (2024.findings-acl)

Challenge:	Existing datasets for instruction-following are monolingual and centered on English . existing data are unable to capture linguistic and cultural subtle differences .
Approach:	They propose an extension of IFEval to a localized multilingual version called Marco-Bench-MIF . their benchmark addresses linguistic constraints and cultural references via translation and verification .
Outcome:	The proposed extension of IFEval to a localized multilingual version covers 30 languages with varying levels of localization.

Copied to clipboard

Xu Huang, Zhirui Zhang, Xiang Geng, Yichao Du, Jiajun Chen, Shujian Huang

Improved Pseudo Data for Machine Translation Quality Estimation with Constrained Beam Search (2023.emnlp-main)

Challenge:	Recent studies have shown that Large Language Models (LLMs) can be used as translation evaluators.
Approach:	They propose to use both coarse-grained and fine-grounded prompts to discern the utility of source versus reference data in machine translation evaluation tasks.
Outcome:	The proposed model can be used to evaluate translations in multiple languages.

Copied to clipboard

Xiang Geng, Yu Zhang, Zhejian Lai, Shuaijie She, Wei Zou, Shimin Tao, Hao Yang, Jiajun Chen, Shujian Huang

Text Generation

Challenge:	evaluating the quality of machine translation outputs becomes increasingly essential with the rapid development of machine language (MT).
Approach:	They propose to generate pseudo data using the MT model with constrained beam search (CBSQE) they propose to preserve the reference parts with high MT probabilities as correct translations .
Outcome:	The proposed model outperforms strong baselines in both supervised and unsupervised settings.

MEPT: Mixture of Expert Prompt Tuning as a Manifold Mapper (2025.emnlp-main)

Copied to clipboard

Runjia Zeng, Guangyan Sun, Qifan Wang, Tong Geng, Sohail Dianat, Xiaotian Han, Raghuveer Rao, Xueling Zhang, Cheng Han, Lifu Huang, Dongfang Liu

Multimodal Dialogue Response Generation (2022.acl-long)

Challenge:	Empirical evaluations show that Mixture of Expert Prompt Tuning outperforms state-of-the-art parameter efficient baselines on SuperGLUE.
Approach:	They propose a pretrain-then-fine-tune paradigm for manifold mapping using multiple prompt experts.
Outcome:	Empirical results show that the proposed approach outperforms state-of-the-art methods on SuperGLUE while reducing activated prompts by 79.25%.

Copied to clipboard

Qingfeng Sun, Yujing Wang, Can Xu, Kai Zheng, Yaming Yang, Huang Hu, Fei Xu, Jessica Zhang, Xiubo Geng, Daxin Jiang

Dialogue Systems & Conversational Agents

Challenge:	Existing studies focus on multimodal dialogue models but neglect generation methods.
Approach:	They propose a multimodal dialogue response generation task which requires multimodal dialogs containing both texts and images which are difficult to obtain.
Outcome:	Experiments show that the proposed model can generate informative text and high-resolution image responses.

MeMoTune: A Measure and Moment-Driven Fine-Tuning Framework for Quantized Large Language Models (2025.findings-acl)

Copied to clipboard

Yun Zhang, Xue Geng, Lizi Liao, Jintong Sun, Minghe Yu, Ge Yu

From Remembering to Metacognition: Do Existing Benchmarks Accurately Evaluate LLMs? (2025.findings-emnlp)

Challenge:	Existing methods combine quantization with parameter-efficient fine-tuning but fail to meet practical performance requirements.
Approach:	They propose a measure and moment approach to optimize objective function for superior fine-tuning results by scaling the update process through a gradient.
Outcome:	The proposed framework outperforms state-of-the-art methods on tasks like text generation, summarization, and understanding.

Copied to clipboard

Geng Zhang, Yizhou Ying, Sihang Jiang, Jiaqing Liang, Guanglei Yue, Yifei Fu, Hailin Hu, Yanghua Xiao

Diffusion vs. Autoregressive Language Models: A Text Embedding Perspective (2025.emnlp-main)

Challenge:	Existing benchmark datasets focus on low-level cognitive tasks while providing limited coverage of higher-level reasoning skills.
Approach:	They analyze the cognitive depth of popular LLM benchmarks using Bloom’s Taxonomy to evaluate both the cognitive and knowledge dimensions.
Outcome:	The results show that incorporating higher-level cognitive instructions into the current instruction fine-tuning process improves model performance.

Copied to clipboard

Siyue Zhang, Yilun Zhao, Liyuan Geng, Arman Cohan, Anh Tuan Luu, Chen Zhao

Detecting Knowledge Boundary of Vision Large Language Models by Sampling-Based Inference (2025.emnlp-main)

Challenge:	Large language model (LLM)-based embedding models surpass BERT and T5 on general-purpose text embeddable tasks.
Approach:	They propose to adopt diffusion language models for text embeddings to overcome limitations in unidirectional attention used during autoregressive pre-training.
Outcome:	The proposed model outperforms the existing LLM-based embedding model on reasoning tasks by 20% and 2% on traditional embeddable benchmarks.

Copied to clipboard

Zhuo Chen, Xinyu Wang, Yong Jiang, Zhen Zhang, Xinyu Geng, Pengjun Xie, Fei Huang, Kewei Tu

KBioXLM: A Knowledge-anchored Biomedical Multilingual Pretrained Language Model (2023.findings-emnlp)

Challenge:	Existing methods to detect the knowledge boundary of Vision Large Language Models (VLLMs) are expensive and require indiscriminate retrieval to address questions that require real-time information or are knowledge-intensive.
Approach:	They propose a method that fine-tunes a VLLM on an automatically constructed dataset for boundary identification.
Outcome:	The proposed method reduces indiscriminate retrieval while maintaining or improving the performance of a VLLM on an automatically constructed dataset.

Copied to clipboard

Lei Geng, Xu Yan, Ziqiang Cao, Juntao Li, Wenjie Li, Sujian Li, Xinjie Zhou, Yang Yang, Jun Zhang

Modeling Event-Pair Relations in External Knowledge Graphs for Script Reasoning (2021.findings-acl)

Challenge:	Existing models for multilingual biomedical training are monolingual, resulting in limited cross-lingual capability.
Approach:	They propose a model that transforms a multilingual biomedical corpus into a biomedically domain using a knowledge-anchored approach.
Outcome:	The proposed model outperforms monolingual and multilingual models in cross-lingual scenarios.

Copied to clipboard

Yucheng Zhou, Xiubo Geng, Tao Shen, Jian Pei, Wenqiang Zhang, Daxin Jiang

Knowledge Representation

Challenge:	Existing methods focus on graph triples with event overlap, but ignore more supportive triples . Script reasoning relies on understanding the relationship between two events .
Approach:	They propose a model to learn the inferential relations between events from the whole eventuality KG . they propose 'script adapter' to extend the model to infer the associated relations between an event chain and a subsequent event candidate.
Outcome:	The proposed model is compared with baselines using external KG or not on a script reasoning task.

Probing Graph Decomposition for Argument Pair Extraction (2023.findings-acl)

Copied to clipboard

Yang Sun, Bin Liang, Jianzhu Bao, Yice Zhang, Geng Tu, Min Yang, Ruifeng Xu

Structured Data in NLP

Challenge:	Argument pair extraction (APE) aims to extract interactive argument pairs from two passages within a discussion.
Approach:	They propose a method to extract interactive argument pairs from two passages . they propose to decompose the probing graph into four sub-graphs based on inter- and intra-passage perspectives .
Outcome:	The proposed method improves on strong baselines on two benchmark datasets.

Shaping the Safety Boundaries: Understanding and Defending Against Jailbreaks in Large Language Models (2025.acl-long)

Copied to clipboard

Lang Gao, Jiahui Geng, Xiangliang Zhang, Preslav Nakov, Xiuying Chen

SafeConf: A Confidence-Calibrated Safety Self-Evaluation Method for Large Language Models (2025.findings-emnlp)

Challenge:	Understanding how jailbreaking works remains limited, hindering the development of effective defense strategies.
Approach:	They propose a new mechanism that adaptively constrains activations within the safety boundary and propose 'Activation Boundary Defense' to enhance its effectiveness.
Outcome:	The proposed defense achieves an average Defense Success Rate (DSR) of over 98% against various jailbreak attacks, with less than 2% impact on the model’s general capabilities.

Copied to clipboard

Bo Zhang, Cong Gao, Linkang Yang, Bingxu Han, Minghao Hu, Zhunchen Luo, Guotong Geng, Xiaoying Bai, Jun Zhang, Wen Yao, Zhong Wang

Multilingual Federated Low-Rank Adaptation for Collaborative Content Anomaly Detection across Multilingual Social Media Participants (2025.emnlp-main)

Challenge:	Large language models (LLMs) have many advantages but they also pose significant safety risks.
Approach:	They propose a method to enhance the safety self-evaluation capability of LLMs . they perform semantic mutations on the original safety evaluation questions .
Outcome:	The proposed method improves safety self-evaluation accuracy by 5.86% and 7.79% over baseline methods on Chinese and English datasets.

Copied to clipboard

Jiaxin Li, Geng Zhao, Xiaoci Zhang

CoreEval: Automatically Building Contamination-Resilient Datasets with Real-World Knowledge toward Reliable LLM Evaluation (2025.acl-long)

Challenge:	Recent developments in multilingual social media platforms (SNS) exacerbate new challenges in SNS content anomaly detection due to data islands and linguistic imbalance.
Approach:	They propose a multilingual Federated LoRA based on SVD-based language-specific disentanglement of LoRA blocks and a local orthogonal tuning strategy to detect content anomalies.
Outcome:	The proposed solution is superior in multilingual content anomaly detection while reducing multilingual knowledge conflicts and communication rounds.

Copied to clipboard

Jingqian Zhao, Bingbing Wang, Geng Tu, Yice Zhang, Qianlong Wang, Bin Liang, Jing Li, Ruifeng Xu

Improving Zero-Shot Cross-lingual Transfer for Multilingual Question Answering over Knowledge Graph (2021.naacl-main)

Challenge:	Publicly available datasets can be used to evaluate performance of large language models . however, contamination of test data can artificially inflate model performance .
Approach:	They propose a Contamination-resilient Evaluation strategy that updates data with real-world knowledge.
Outcome:	The proposed evaluation strategy can be used to update datasets with real-world knowledge.

Copied to clipboard

Yucheng Zhou, Xiubo Geng, Tao Shen, Wenqiang Zhang, Daxin Jiang

VIP5: Towards Multimodal Foundation Models for Recommendation (2023.findings-emnlp)

Challenge:	Existing approaches to solve multilingual question answering over knowledge graph (KGQA) use of bilingual lexicon induction to map training questions into those in target language circumvents language inconsistency .
Approach:	They propose to use bilingual lexicon induction to map training questions in source and target languages as augmented training data to minimize syntax-disorder.
Outcome:	The proposed model narrows the gap in zero-shot cross-lingual transfer between source and target languages.

Copied to clipboard

Shijie Geng, Juntao Tan, Shuchang Liu, Zuohui Fu, Yongfeng Zhang

Multimodality

Challenge:	Recent advances in foundation models have impeded the ability for these fields to benefit from each other’s advancements.
Approach:	They propose to use a multimodal foundation model to unify various modalities and recommendation tasks under the P5 recommendation paradigm to implement personalized prompts.
Outcome:	The proposed model will unify visual, textual, and personalization modalities under the P5 recommendation paradigm and will improve recommendation performance and efficiency.

Scaling Behaviors of LLM Reinforcement Learning Post-Training: An Empirical Study in Mathematical Reasoning (2026.acl-long)

Copied to clipboard

Zelin Tan, Hejia Geng, Xiaohang Yu, Mulei Zhang, Guancheng Wan, Yifan Zhou, Qiang He, Xiangyuan Xue, Heng Zhou, Yutao Fan, Zhong-Zhi Li, Zaibin Zhang, Guibin Zhang, Chen Zhang, Zhenfei Yin, Philip Torr, Lei Bai

The Paradox of Outcome Optimization: A Causal Information-Theoretic Bound on Reasoning Shortcuts in LLMs (2026.acl-long)

Challenge:	elucidating scaling laws for large language models (LLMs) during pre-training remains unexplored.
Approach:	They characterize how model scale, data, and compute interact during pre-training . they find that large models consistently demonstrate superior compute and data efficiency .
Outcome:	The proposed scaling laws offer practical guidance for scaling reasoning capabilities through reinforcement learning post-training.

Copied to clipboard

Zihan Chen, Yiming Zhang, Wenxiang Geng, Zenghui Ding, Yining Sun

Dynamic Evil Score-Guided Decoding: An Efficient Decoding Framework For Red-Team Model (2025.findings-acl)

Challenge:	Large Language Models (LLMs) aligned via outcome-based Reinforcement Learning (RL) exhibit a critical failure mode: they exhibit brittle reasoning capabilities on out-of-distribution tasks.
Approach:	They propose a framework bridging Structural Causal Models and the Information Bottleneck principle to explain this paradox.
Outcome:	The proposed framework bridges the framework between SCM and IB principles to explain the problem.

Copied to clipboard

Cong Gao, Bo Zhang, Linkang Yang, Minghao Hu, Zhunchen Luo, Xiaoying Bai, Guotong Geng, Jun Zhang, Yunhua Xue