Paper 'MATH-Perturb: Benchmarking LLMs' Math Reasoning Abilities against Hard Perturbations' accepted at ICML 2025 (scheduled talk in May 2025)
Paper 'CRISPR-GPT: An LLM Agent for Automated Design of Gene-Editing Experiments' to appear in Nature Biomedical Engineering (scheduled talk at INFORMS 2024 in October 2024)
Paper 'Assessing the Brittleness of Safety Alignment via Pruning and Low-Rank Modifications' received Best Paper Award at ICML 2024 and ICLR Secure and Trustworthy LLMs 2024
Paper 'Visual Adversarial Examples Jailbreak Large Language Models' accepted as Oral presentation at AAAI 2024
Developed SpecDec++, an efficient decoding algorithm, published at COLM 2025 and ICML 2024 ES-FoMo workshop