Xinfeng Li

Defect classification using the pulsed alternating current field measurement technique based on the Smoothed Pseudo Wigner-Ville distribution and principal component analysis

Qingxiao Kong, Lilong Lin, Xinfeng Li, Shuwei Pan et autres

Introduction Pulsed Alternating Current Field Measurement (PACFM) has been proven to offer significant advantages in detecting surface and subsurface defects in structural components, and has been successfully applied to nondestructive testing of multilayer structures such as aircraft wings. Although pulsed excitation response …

cn (code pays fourni par la source)

Cybersecurity vulnerabilities in IoT devices

Chen Yan, Xiaoyu Ji, Qinhong Jiang, K Wang et autres

cn (code pays fourni par la source)

ENCHTABLE: Unified Safety Alignment Transfer in Fine-Tuned Large Language Models

Jialin Wu, Kaiwen Li, Zhicong Huang, Xinfeng Li et autres

Many machine learning models are fine-tuned from large language models (LLMs) to achieve high performance in specialized domains like code generation, biomedical analysis, and mathematical problem solving. However, this fine-tuning process often introduces a critical vulnerability: the systematic degradation of safety alignment, …

cn, sg (code pays fourni par la source)

WebCloak: Characterizing and Mitigating Threats From LLM-Driven Web Agents as Intelligent Scrapers

Xinfeng Li, Tianze Qiu, Yingbin Jin, Lixu Wang et autres

sg, hk, us (code pays fourni par la source)

EmoRAG: Evaluating RAG Robustness to Symbolic Perturbations

Xin Zhou, Xinfeng Li, Yinan Peng, Ming Sheng Xu et autres

Retrieval-Augmented Generation (RAG) systems are increasingly central to robust AI, enhancing large language model (LLM) faithfulness by incorporating external knowledge. However, our study unveils a critical, overlooked vulnerability: their profound susceptibility to subtle symbolic perturbations, particularly through near-imperceptible emotional icons (e.g., "(@_@)") …

cn, us (code pays fourni par la source)

Dynamic Generation of Multi LLM Agents Communication Topologies with Graph Diffusion Models

Eric Hanchen Jiang, Mengting Li, Xinfeng Li, Yizhou Sun et autres

Eric Hanchen Jiang, Levina Li, Frank Wan, Xiao Liang, Sophia Yin, Yuchen Wu, Xinfeng Li, Yizhou Sun, Wei Wang, Kai-Wei Chang, Ying Nian Wu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Mitigating Over-Refusal in Aligned Large Language Models via Inference-Time Activation Energy

Eric Hanchen Jiang, Weixuan Ou, Run Liu, Shaoning Pang et autres

Eric Hanchen Jiang, Weixuan Ou, Run Liu, Shengyuan Pang, Guancheng Wan, Ranjie Duan, Wei Dong, Kai-Wei Chang, XiaoFeng Wang, Ying Nian Wu, Xinfeng Li. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

us, cn (code pays fourni par la source)

AP60: A Taxonomy-Guided Benchmark Dataset for Fine-Grained Pest Recognition with Feature-Level Confusion Analysis

Xianfeng Zhou, Shaogang Lei, Xinfeng Li, Zhaojie Zhang et autres

Accurate recognition of visually similar pest species remains a major challenge in agricultural vision, given that existing datasets often lack sufficient taxonomic structure, confusable categories, and quantitative analysis of class-level visual difficulty. To address these limitations, we present AP60, a taxonomy-guided benchmark …

cn (code pays fourni par la source)

CentaurEval: Benchmarking Human-in-the-Loop Value in Agentic Coding

Cheng Ni, Yiran Wang, Yingbin Jin, Xinfeng Li et autres

LLM-powered coding agents are reshaping the development paradigm. However, existing evaluation systems, neither traditional tests for humans nor benchmarks for LLMs, fail to capture this shift, excluding problems that require both human reasoning to guide solutions and AI efficiency for implementation. We …

The Eye of Sherlock Holmes: Uncovering User Private Attribute Profiling via Vision-Language Model Agentic Framework

Feiran Liu, Yinan Peng, Xinfeng Li, Lixu Wang et autres

cn, sg, us (code pays fourni par la source)

Critical Information Only: A Content Privacy-Preserving Framework for Detecting Audio Deepfakes

Xinfeng Li, Kai Li, Chang Zeng, Xiaoyu Ji et autres

Text-to-Speech (TTS) and Voice Conversion (VC) models have exhibited remarkable performance in generating realistic and natural audio. However, their dark side, audio deepfake poses a significant threat to both society and individuals. Existing countermeasures largely focus on determining the genuineness of speech …

sg, cn, jp (code pays fourni par la source)

A Vision for Access Control in LLM Agent Systems

Dong Huang, Jiakui Li, Hongyi Cai, Xinfeng Li et autres

sg, cn, my (code pays fourni par la source)