Assistant Professor at Beijing Key Laboratory of AI Safety and Superalignment and BrainCog Lab
Ph.D., CAS Institute of Automation (with Prof. Yi Zeng)
B.Sc., School of Math & Stats, Xidian University
I currently serve as a reviewer for several leading conferences and journals, including IJCAI, NeurIPS, ICLR, Neural Networks, Neurocomputing, Pattern Recognition, and IEEE journals such as TNNLS, TIP, and TETCI.
PANDA (Platform on Attack aNd Defense Assessment) is an AI safety evaluation platform that benchmarks 30+ leading open- and closed-source LLMs — including GPT-4o, DeepSeek-V3, Claude 3, LLaMA 3, Qwen 2.5, Mistral, Gemma, and Grok. The platform currently supports 15 attack algorithms and 12 defense mechanisms, enabling systematic assessment of LLM robustness and alignment safety.
Below is a visualization of attack success rates (ASR) of various domestic and international foundation models on the Jailbreak Benchmark dataset under different attack strategies.
Spiking Transformer Benchmark is a unified framework for reproducing and evaluating existing spiking Transformer models. It provides standardized interfaces for classification, detection, and segmentation across both static and event-based datasets. The benchmark aims to offer fair, extensible, and modular comparisons. Future extensions will support audio and text tasks, enabling broader exploration of Spiking Transformers across modalities.
Brain-inspired Cognitive Intelligence Engine (BrainCog) is a brain-inspired neural network based platform for realizing Brain-inspired Artificial Intelligence, and simulating the cognitive brains of different animal species at multiple scales. The ultimate goal and long term efforts of BrainCog is to provide a comprehensive theory and systems to decode the mechanisms and principles of human intelligence and its evolution, and develop artificial brains for brain-inspired conscious living becomings for the future human-AI symbiotic society.
Project Site