Research Manuscript: Do Not Forget the Software: Bare Metal Neural Acceleration is no fun without it.
Event TypeResearch Manuscript
AI/ML Design: Circuits and Architecture
TimeTuesday, July 12th3:30pm - 5:30pm PDT
Location3002, Level 3
DescriptionThis session presents various design automation and optimization frameworks for energy efficient and high-performance neural network applications. The papers of this session cover among others, techniques used during building the models, such as conversion aware training and model compression optimization methods. Further, various design automation approaches for ReRAM-based in-memory computing and mapping of weights in non-volatile memory for in-memory accelerators, are presented, some emphasizing their usefulness in multi-task deep neural networks.