Published in IEEE BigData 2021, 2021
An attention-based U-Net with residual feature extraction for improved 2D medical image segmentation.
Recommended citation: Zhang, X., Feng, Z., Zhong, T., Shen, S., Zhang, R., Zhou, L., Zhang, B., & Wang, W. (2021). DRA U-Net: An Attention based U-Net Framework for 2D Medical Image Segmentation.
Download Paper
Published in IJCAI 2024, 2024
The first Spiking Transformer enhancement method from the perspective of temporal enhancement.
Recommended citation: Shen, S., Zhao, D., Shen, G., & Zeng, Y. (2024). TIM: an efficient temporal interaction module for spiking transformer.
Download Paper
Published in Tiny Papers @ ICLR 2024, 2024
A binary attention design for Spiking Transformers that improves sparsity and reduces computation.
Recommended citation: Shen, G., Zhao, D., Shen, S., & Zeng, Y. (2024). Enhancing Spiking Transformers with Binary Attention Mechanisms.
Download Paper
Published in arXiv, 2024
A time-cell-inspired temporal codebook for improving image generation with spiking neural networks.
Recommended citation: Feng, L., Zhao, D., Shen, S., Dong, Y., Shen, G., & Zeng, Y. (2024). Time Cell Inspired Temporal Codebook in Spiking Neural Networks for Enhanced Image Generation.
Download Paper
Published in arXiv, 2025
A spiking diffusion model that introduces adaptive lateral connections for more expressive and efficient generation.
Recommended citation: Feng, L., Zhao, D., Shen, S., & Zeng, Y. (2025). Biologically Inspired Spiking Diffusion Model with Adaptive Lateral Selection Mechanism.
Download Paper
Published in arXiv, 2025
We introduce PandaGuard and PandaBench, a unified, reproducible framework and benchmark for systematically evaluating LLM jailbreak attacks, defenses, and judges, revealing that no single defense is universally optimal and that judge disagreement significantly affects safety assessments.
Recommended citation: Shen, G., Zhao, D., Feng, L., He, X., Wang, J., Shen, S., ... & Zeng, Y. (2025). PANDAGUARD: Systematic Evaluation of LLM Safety against Jailbreaking Attacks.
Download Paper
Published in NeurIPS 2025, 2025
The first Spiking Transformer evaluating models on various datasets and neuron encoding shcemas via unified method.
Recommended citation: Shen, S., Zhao, D., Feng, L., Yue, Z., Li, J., Li, T., ... & Zeng, Y. (2025). STEP: A Unified Spiking Transformer Evaluation Platform for Fair and Reproducible Benchmarking.
Download Paper
Published in ICML 2026, 2026
The first brain-inspired bi-directional temporal enhancement on Spiking Transformers.
Recommended citation: Shen, S., Lv, M., Han, B., Zhao, D., Shen, G., Zhao, F., & Zeng, Y. (2026). TEFormer: Structured Bidirectional Temporal Enhancement Modeling in Spiking Transformers
Download Paper
Published in arXiv, 2026
A lightweight safety-aware decoding method that improves LLM safety with a single-neuron gate and low training cost.
Recommended citation: Shen, S., Lv, M., Shen, H., Wu, J., Wang, B., Yang, Z., Shen, G., Zhao, D., Zhao, F., & Zeng, Y. (2026). Light Alignment Improves LLM Safety via Model Self-Reflection with a Single Neuron.
Download Paper