In this research project, students will investigate cutting-edge challenges at the intersection of machine learning, security, and system robustness. Participants may choose from several current research topics spanning model alignment, AI safety, and hardware-level attacks on modern accelerators. After selecting a topic, students will join an active research group and receive close guidance from an experienced supervisor throughout the project. Possible research directions include:

• Robust Safety Alignment Against Neuron-Pruning Jailbreaks
• GPU Rowhammer Attacks on Mixture-of-Expert (MoE) LLMs
• Human-Guided Generative AI for High-Performance Hardware Fuzzing * Exploiting and Mitigating Vulnerabilities in Human-in-the-Loop Self-Learning LLM Agents
• RAG Poisoning for LLMs Students will gain hands-on experience designing experiments, analyzing security vulnerabilities, developing mitigation strategies, and communicating research findings. This project is ideal for students interested in AI safety, adversarial machine learning, systems security, or large-scale neural network architectures.