Research 

My research interests are at the intersection of Software Engineering and Artificial Intelligence (AI). I am particularly interested in:

1. Improving the performance of DL and LLM-based systems on the fly through input refinement, adaptation, and uncertainty-aware validation techniques across classification and generation tasks in code, vision, and NLP domains;

2. GPU testing for automated bug detection, focusing on numerical bugs in CUDA, Triton AI, and RISC-V programs using program analysis and test generation;

3. Efficiency robustness of dynamic deep learning systems (DDLSs) by studying and designing efficiency attacks and defenses under varying threat models, with emphasis on sparsity, Mixture-of-Experts architectures, and adaptive computation.