I am Sicheng Shen (沈思成), a Ph.D. student at the Institute of Automation, Chinese Academy of Sciences, affiliated with the Beijing Zhongguancun Academy.

My research focuses on brain-inspired artificial intelligence, spiking neural networks, LLM safety and alignment, and embodied / vision-language-action safety. I am interested in developing learning systems that are efficient, biologically motivated, and reliable in open-world environments.

My recent work includes temporal modeling in Spiking Transformers, systematic evaluation of LLM jailbreak safety, lightweight alignment methods for large language models, and trustworthy alignment for vision-language-action systems.

Links: Google Scholar · GitHub · Semantic Scholar

Research Interests

• Brain-inspired artificial intelligence
• Spiking neural networks and Spiking Transformers
• LLM safety, jailbreak evaluation, and alignment
• Vision-language-action models and embodied AI safety

Selected Publications

• TEFormer: Structured Bidirectional Temporal Enhancement Modeling in Spiking Transformers
• STEP: A Unified Spiking Transformer Evaluation Platform for Fair and Reproducible Benchmarking
• PANDAGUARD: Systematic Evaluation of LLM Safety against Jailbreaking Attacks
• TIM: an efficient temporal interaction module for spiking transformer
• Light Alignment Improves LLM Safety via Model Self-Reflection with a Single Neuron

Education

• Institute of Automation, CAS & UCAS, direct Ph.D. track, 2024.09 - Present
  GPA: 3.9 / 4.0
• Beijing University of Posts & Telecom., B.Eng. in IOT, 2020.09 - 2024.06
  GPA: 3.8 / 4.0, Rank: 2 / 190

Research and Projects

• Brain-inspired AI, 2023.08 - Present Working on brain-inspired learning systems with a focus on temporal modeling, architecture design, optimization, benchmarking, and downstream applications of spiking neural networks and Spiking Transformers.
• Panda-Guard, 2024.12 - 2025.04
  Participated in the development of a unified benchmark for LLM jailbreak attacks, defenses, and judges. Related links: code, paper.
• Lightweight LLM Alignment, 2025.09 - Present
  Exploring training-efficient and inference-efficient lightweight alignment strategies for large language models, with current work targeting ICML 2026.
• Trustworthy VLA Alignment for Open-World Scenarios, 2024.01 - Present
  Studying reliable alignment and evaluation for VLA systems, with an emphasis on benchmarking and post-training methods for open-world deployment.

Selected Honors

• Queen Mary Prize, Queen Mary University of London, 2024
• Beijing Outstanding Undergraduate Thesis, 2024
• Beijing Normal University Outstanding Graduate, 2024
• National Third Prize, National English Competition for College Students (NECCS), 2021
• Iwate friendship ambassador, Japan 2015