System Software for Machine Learning at the Edge - tinyML Asia 2020
Explore system software approaches for machine learning at the edge in this tinyML Asia 2020 conference talk. Delve into the challenges and opportunities of deploying ML on mobile and IoT devices, focusing on compiler and runtime software strategies to maximize hardware potential. Learn about deep neural network optimizations, workload partitioning, and voltage-frequency scaling techniques for orchestrating on-chip compute resources. Discover how these methods can achieve low-power, real-time edge ML performance across various applications, from gesture recognition to fitness tracking. Gain insights into the evolving landscape of heterogeneous system-on-chips, TensorFlow Lite architecture, and the role of compilers in addressing fundamental problems in edge computing.