Machine Learning Systems

Through exploiting sparsity, quantization, and co-designing the electrical and optical hardware, our work aims to improve the performance, energy efficiency, and robustness (against adversarial attacks, soft errors, and permanent faults) of modern deep learning systems.