Research

As learning model sizes growing faster than hardware efficiency improves, conventional computing architectures are becoming increasingly unsustainable. Rather than relying on ever-larger centralized hardware, I envision next-generation embedded systems that perform inference locally, operate on energy harvested from the environment, and remain secured.

Our research design inference algorithms targeting both von Neumann and non-von Neumann architectures. Specifically, we study neural network and hyper-dimensonal computing from a systems perspective, optimizing four aspects in resource-constrained environments: (1) memory, (2) computation, (3) energy efficiency, and (4) privacy.

Our research has appeared in venues such as ICCAD, DATE, IEEE TCAD, and ACM TODAES. I also serve on the Technical Program Committees (TPC) of DAC and MLSys and reviewer for IEEE TCAD.

Research Interests:

• Intelligence on and beyond the Edge
• Energy-Efficient Inference
• Privacy Preserving Inference