Wei Shen

I'm a master student at Wuhan University, under the supervision of Prof. Mang Ye. Before that, I obtained my B.S. degree from Wuhan University in 2023.

My research interests lie in Trustworthy AI, with a focus on Federated Learning and Large Language Models. I am committed to developing practical, generalizable algorithms that enhance the robustness and explainability of ML systems, making them controllable.

Email / Google Scholar / Github

News

[05/2026] Our work about deceptive behavior in LLM CoTs (DecepChain) has been accepted by ICML 2026. See you in Seoul!
[09/2025] Our benchmark about realistic evaluations in VFL has been accepted by NeurIPS 2025 as spotlight presentation. See you in San Diego!
[02/2025] Our paper about adressing limited aligned samples in VFL has been accepted by IEEE Transactions on Mobile Computing (TMC).
[12/2024] Our paper about backdoor attacks in VFL has been accepted at AAAI 2025. See you in Philadelphia!
[07/2024] Our paper about handling oversmoothing in GNN has been accepted at ACM MM 2024. See you in Melbourne!

Selected Publications

	DecepChain: Inducing Deceptive Reasoning in Large Language Models [paper] [project] Wei Shen⁺, Han Wang⁺, Haoyu Li⁺, Huan Zhang Forty-Third International Conference on Machine Learning (ICML), 2026 We investigate the intriguing deceptive behavior in existing LLMs where CoTs are incorrect but look plausible for human users.
	MARS-VFL: A Unified Benchmark for Vertical Federated Learning with Realistic Evaluation (Spotlight) [paper] [codes] Wei Shen, Weiqi Liu, Mingde Chen, Wenke Huang, Mang Ye The 39th Conference on Neural Information Processing Systems (NeurIPS), 2025 We propose a unified benchmark for realistic VFL evaluation that integrates data from practical applications.
	Label-Free Backdoor Attacks in Vertical Federated Learning [paper] [codes] Wei Shen⁺, Wenke Huang⁺, Guancheng Wan, Mang Ye The 39th Annual AAAI Conference on Artificial Intelligence (AAAI), 2025 We propose a backdoor paradigm for VFL that operates without explicit label information.
	Resisting Over-Smoothing in Graph Neural Networks via Dual-Dimensional Decoupling [paper] [codes] Wei Shen, Mang Ye, Wenke Huang ACM Multimedia (ACM MM), 2024 We address the oversmoothing issue in GNNs by leveraging both instance-level and dimension-level cues.

Awards

National Scholarship, 2025.11.
NeurIPS Financial Aid Award, 2025.10
Tencent Scholarship (Special Prize), 2025.10
DiDi Scholarship, 2025.10
National Encouragement Scholarship, 2022.09

Service

Conference Reviewer: ICLR 2025/2026, NeurIPS 2026, CVPR 2024/2025/2026, AAAI 2026.

Thanks for the template from Jon Barron.