My research interests include Systems for ML, HW/ML Model Co-Design, Hardware-Aware ML, and recently Reinforcement Learning
My research focuses on designing scalable and efficient systems and ML models using HW/ML model co-design techniques to achieve the best of both worlds. Currently, I have been working on quantization-aware DNN accelerator and model co-exploration through architecture-level modeling and efficient design space exploration. Recently, I have been working on scalable and efficient reinforcement learning training on CPU-GPU systems. Additionally, my previous work has explored how to utilize emerging non-volatile memories in GPU architectures for DL workloads.