SecProbe is an adaptive evaluation framework for coding agents on cybersecurity vulnerabilities. It combines Item Response Theory with on-demand synthesis of repository-scale vulnerability-repair tasks, using observed performance to estimate agent ability and select or generate informative tasks. Across 353 tasks spanning six languages and 151 CWE types, nine frontier models reach at most 28.33% success, and SecProbe matches baseline ability estimates with up to 29.5% fewer tasks.
@article{luo2026secprobe,title={SecProbe: Adaptive Evaluation of Coding Agents on Cybersecurity Vulnerabilities},author={Luo, Xiaonan and Huang, Yue and Guo, Kehan and He, Ping and Zou, Chuan and Gao, Chujie and Li, Lichi and Ma, Yuchen and Xu, Zhangchen and Chen, Zichen and Han, Yufei and Zhang, Xiangliang},journal={arXiv preprint arXiv:2609.33763},year={2026},}
arXiv
Learning from Synthetic Data without Model Collapse in Iterative Instruction Tuning
Xiaonan Luo, Yue Huang, Kehan Guo, and 4 more authors
We find that model collapse in iterative instruction tuning is a polarization of competence: using DINA-based skill diagnosis, we show that synthetic training can reinforce strong skills while degrading weakly supported ones. We propose KITE, a two-stage synthetic-data framework that couples weakness-guided candidate generation with boundary-aware uncertainty curation, enabling more stable multi-generation improvement than strong synthetic-data baselines across reasoning benchmarks.
@article{luo2026kite,title={Learning from Synthetic Data without Model Collapse in Iterative Instruction Tuning},author={Luo, Xiaonan and Huang, Yue and Guo, Kehan and He, Ping and Zou, Chuan and Hua, Ting and Zhang, Xiangliang},journal={arXiv preprint arXiv:2607.17043},year={2026},}
arXiv
MemoHarness: Agent Harnesses That Learn from Experience
Yue Huang, Wenjie Wang, Han Bao, and 7 more authors
@article{huang2026memoharness,title={MemoHarness: Agent Harnesses That Learn from Experience},author={Huang, Yue and Wang, Wenjie and Bao, Han and Ma, Yuchen and Luo, Xiaonan and Nian, Yi and Zhuang, Haomin and Liu, Zheyuan and Zhao, Yue and Zhang, Xiangliang},journal={arXiv preprint arXiv:2607.14159},year={2026},}
arXiv
UXBench: Measuring the Actionability of LLM-Generated UX Critiques
Wenjie Wang, Yue Huang, Zipeng Ling, and 11 more authors
@article{wang2026uxbench,title={UXBench: Measuring the Actionability of LLM-Generated UX Critiques},author={Wang, Wenjie and Huang, Yue and Ling, Zipeng and Bao, Han and Hua, Hang and Luo, Xiaonan and Jiang, Yu and Du, Shiyi and Hao, Yuexing and Li, Xiaomin and Ma, Yuchen and Wang, Dianzhuo and Ye, Yanfang and Zhang, Xiangliang},journal={arXiv preprint arXiv:2606.16262},year={2026},}
EMNLP
Right Answers for the Right Reasons: Coherence-Enhanced Policy Optimization for Large Language Models
Xiangqi Wang, Yue Huang, Yujun Zhou, and 3 more authors
In Conference on Empirical Methods in Natural Language Processing, 2026
@inproceedings{wang2026cepo,title={Right Answers for the Right Reasons: Coherence-Enhanced Policy Optimization for Large Language Models},author={Wang, Xiangqi and Huang, Yue and Zhou, Yujun and Luo, Xiaonan and Guo, Kehan and Zhang, Xiangliang},booktitle={Conference on Empirical Methods in Natural Language Processing},year={2026},}
KDD
Synthetic Interaction Data for Scalable Personalization in Large Language Models
Yuchen Ma, Yue Huang, Wenjie Wang, and 3 more authors
In ACM SIGKDD Conference on Knowledge Discovery and Data Mining, Datasets and Benchmarks Track, 2026
@inproceedings{ma2026synthetic,title={Synthetic Interaction Data for Scalable Personalization in Large Language Models},author={Ma, Yuchen and Huang, Yue and Wang, Wenjie and Luo, Xiaonan and Zhang, Xiangliang and Feuerriegel, Stefan},booktitle={ACM SIGKDD Conference on Knowledge Discovery and Data Mining, Datasets and Benchmarks Track},year={2026},}
arXiv
Reliable Control-Point Selection for Steering Reasoning in Large Language Models
Haomin Zhuang, Hojun Yoo, Xiaonan Luo, and 2 more authors
@article{zhuang2026controlpoint,title={Reliable Control-Point Selection for Steering Reasoning in Large Language Models},author={Zhuang, Haomin and Yoo, Hojun and Luo, Xiaonan and Guo, Kehan and Zhang, Xiangliang},journal={arXiv preprint arXiv:2604.02113},year={2026},}
ICML-W
Emergent Social Intelligence Risks in Generative Multi-Agent Systems
Yue Huang, Yu Jiang, Wenjie Wang, and 12 more authors
@inproceedings{huang2026emergent,title={Emergent Social Intelligence Risks in Generative Multi-Agent Systems},author={Huang, Yue and Jiang, Yu and Wang, Wenjie and Zhuang, Haomin and Luo, Xiaonan and Ma, Yuchen and Xu, Zhangchen and Chen, Zichen and Moniz, Nuno and Lin, Zinan and Chen, Pin-Yu and Chawla, Nitesh V and Dziri, Nouha and Sun, Huan and Zhang, Xiangliang},booktitle={Trustworthy AI for Good Workshop at ICML},year={2026},}
AAAI
Better Datasets Start From RefineLab: Automatic Optimization for High-Quality Dataset Refinement
Xiaonan Luo*, Yue Huang*, Ping He, and 1 more author
In AAAI Conference on Artificial Intelligence, 2026
RefineLab is a framework that formulates QA dataset refinement as an integer linear program, automatically selecting coverage, difficulty, and factual-consistency edits under a token budget to align datasets with explicit quality targets. Its modular refinement operators yield stable evaluation improvements on benchmarks including MMLU, GSM8K, and HumanEval.
@inproceedings{luo2026refinelab,title={Better Datasets Start From RefineLab: Automatic Optimization for High-Quality Dataset Refinement},author={Luo, Xiaonan and Huang, Yue and He, Ping and Zhang, Xiangliang},booktitle={AAAI Conference on Artificial Intelligence},year={2026},}
2025
NeurIPS
AdaReasoner: Adaptive Reasoning Enables More Flexible Thinking
Xiangqi Wang, Yue Huang, Yanbo Wang, and 4 more authors
In Advances in Neural Information Processing Systems, 2025
AdaReasoner introduces an adaptive reasoning framework that enables large language models to dynamically adjust their reasoning strategies based on problem complexity and context.
@inproceedings{wang2025adareasoner,title={AdaReasoner: Adaptive Reasoning Enables More Flexible Thinking},author={Wang, Xiangqi and Huang, Yue and Wang, Yanbo and Luo, Xiaonan and Guo, Kehan and Zhou, Yujun and Zhang, Xiangliang},booktitle={Advances in Neural Information Processing Systems},year={2025},note={Spotlight},}
NeurIPS
ChemOrch: Empowering LLMs with Chemical Intelligence via Synthetic Instructions
Yue Huang, Zhengzhe Jiang, Xiaonan Luo, and 12 more authors
In Advances in Neural Information Processing Systems, 2025
ChemOrch presents a novel approach to enhancing large language models with specialized chemical knowledge through carefully designed synthetic instructions, enabling more accurate chemical reasoning and prediction.
@inproceedings{huang2025chemorch,title={ChemOrch: Empowering LLMs with Chemical Intelligence via Synthetic Instructions},author={Huang, Yue and Jiang, Zhengzhe and Luo, Xiaonan and Guo, Kehan and Zhuang, Haomin and Zhou, Yujun and Yuan, Zhengqing and Sun, Xiaoqi and Schleinitz, Jules and Wang, Yanbo and Zhang, Shuhao and Surve, Mihir and Chawla, Nitesh V and Wiest, Olaf and Zhang, Xiangliang},booktitle={Advances in Neural Information Processing Systems},year={2025},}
ATC
Torpor: GPU-Enabled Serverless Computing for Low-Latency, Resource-Efficient Inference
Mingyu Yu, Ao Wang, Dong Chen, and 11 more authors
Torpor presents a GPU-enabled serverless computing system that achieves low-latency, resource-efficient inference by intelligently managing GPU memory and computation resources.
@inproceedings{yu2025torpor,title={Torpor: GPU-Enabled Serverless Computing for Low-Latency, Resource-Efficient Inference},author={Yu, Mingyu and Wang, Ao and Chen, Dong and Yu, Haoxuan and Luo, Xiaonan and Li, Zhuzhong and Wang, Wei and Chen, Ruichuan and Nie, Dapeng and Yang, Kaiyuan and Chen, Xiaobing and Liu, Mingyang and Zhang, Yijia and Yang, Mao},booktitle={USENIX Annual Technical Conference},year={2025},}