Zichen Song is an incoming Combined Master–PhD student in Computer Science and Engineering at Sungkyunkwan University (SKKU), South Korea (Fall 2025). His research focuses on security and privacy for large language models—particularly membership inference and data extraction attacks—as well as optimization of (multi)modal LLMs, deep spiking neural networks, and medical AI.
His broader interests include trustworthy AI, federated learning, and efficient training/inference for foundation models. He has experience with Python, C/C++, PyTorch, TensorFlow, and Linux, and enjoys building reproducible research codebases and clean experimental pipelines.
Education
-
Sungkyunkwan University (SKKU), South Korea — Combined Master–PhD, Computer Science and Engineering (incoming)
2025 – Present -
Lanzhou University, China — B.S. in Computer Science
2021 – 2025
Publications
- SAMM: A Selective Attention Sequential Model for EEG-EOG Vigilance Estimation — Zichen Song, Yuxi Tong, Yuxin Wu. *CogSci 2025 (CCF-B), Accept (Poster).
- Mamba-CCA: An Efficient Framework for EEG Emotion Recognition — Zichen Song, Yuxi Tong, Yuxin Wu. *CogSci 2025 (CCF-B), Accept (Abstract); expanded to IEEE TETCI (under review).
- Optimization and Robustness-Informed Membership Inference Attacks for LLMs — Zichen Song, Ming Li, Yao Shu. *ICML 2025 DIG-BUG Workshop, Accept (Long); under review at AAAI 2026.
- Mamba-MSCCA-Net: Efficient Change Detection for Remote Sensing Images — Zichen Song, Sitan Huang. Displays, Accept.
- HSCL-RL: Mitigating Hallucinations in Multimodal Large Language Models — Zichen Song, Sitan Huang. NeurIPS 2024 OWA Workshop, Accept (Poster).
- Pulse transfer learning: multi-area river ammonia nitrogen prediction with limited data — Zichen Song, Boying Nie, Sitan Huang (co-first author). Expert Systems with Applications (IF 7.5), Accept (Mar 2024).
- EM-MIAs: Enhancing Membership Inference Attacks in Large Language Models through Ensemble Modeling — Zichen Song, Sitan Huang, Zhongfeng Kang. ICASSP 2025 (CCF-B), Accept.
- Mutual Information Dropout: Mutual Information Can Be All You Need — Zichen Song, Shan Ma. ICANN 2023 (CCF C), Accept (Oral).
- DNN-based hospital service satisfaction using GCNNs learning — Zichen Song, Shan Ma. IEEE Access (IF 3.9), Accept.