Research Manuscript: Hard(ware) Learning is Cool!!
Event TypeResearch Manuscript
AI/ML Design: System and Platform
TimeThursday, July 14th3:30pm - 5:30pm PDT
Location3002, Level 3
DescriptionMachine learning applications have shifted from general-purpose processors to specialized hardware accelerators to achieve high performance and energy efficiency. This session presents six papers on efficient and scalable hardware implementations of machine learning. The papers range from traditional deep learning acceleration to continual learning as well as reinforcement learning. In order to improve efficiency, these papers propose transformative improvement in techniques such as weight pruning and model partitioning. Improving neural compilation using mathematical embeddings of target hardware and efficient hardware-software co-design approaches are also introduced.