Biography

My research interests focus on Inclusive AI: Make AI more open, robust, widely deployable, and efficient in inference:

• DNN Efficiency for CNNs, Diffusion Models, and LLMs
  • Pruning, architecture search, quantization, token reduction, etc.
  • Achieve real-time inference for DNN models on edge devices
• DNN Robustness
  • Explore various DNN attack methods
  • Investigate defense methods against various DNN attacks
• LLM Openness
  • Train fully open-source LLMs
  • Adopt various LLM post-train techniques, such as SFT, DPO, GRPO, etc.
• Optimization Methods
  • Bi-level optimization, ADMM, Bayesian optimization, second-order optimization, zeroth-order optimization, min-max optimization etc.