Presentation
ML-CGRA: An Integrated Compilation Framework to Enable Efficient Machine Learning Acceleration on CGRAs
DescriptionWith the raising of machine learning, computation efficiency arouses widespread concern. The Coarse-Grained-Reconfigurable Array (CGRA) is a programmable logic device and can achieve excellent computation performance with low energy consumption, which is naturally and perfectly fit for machine learning acceleration. Unfortunately, there is no existing framework to facilitate the deployment and acceleration of different machine learning models on CGRAs. This paper proposes ML-CGRA, an integrated compilation framework to enable efficient machine learning acceleration on CGRAs.
Event Type
Research Manuscript
TimeTuesday, July 11th4:10pm - 4:25pm PDT
Location3006, 3rd Floor
Design
AI/ML System and Platform Design