Watch a 28-minute conference talk from tinyML Asia 2022 exploring hardware-aware model optimization techniques for the Arm Ethos-U65 NPU. Learn how analyzing hardware performance through Arm Virtual Hardware revealed convolution layer latency patterns, and discover how these insights inform parameter optimization. Explore how NetsPresso, a hardware-aware AI optimization platform, leverages device-specific characteristics to adjust structured pruning and filter decomposition parameters. See practical validation of how this hardware-aware approach to model compression achieves improved FLOPs and accuracy while maintaining consistent latency in image classification tasks. Gain valuable insights from Nota Inc's Lead Core Research on optimizing deep learning models for edge devices and lightweight neural networks.
Overview
Syllabus
tinyML Asia 2022 Shinkook Choi: Hardware-aware model optimization in Arm Ethos-U65 NPU
Taught by
EDGE AI FOUNDATION
Related Courses
-
Creating End-to-End TinyML Applications for Ethos-U NPU in the Cloud
-
Tiny but Powerful: Hardware for High Performance, Low Power Machine Learning
-
PyNetsPresso and LaunchX: An Integrated Toolchain for Hardware-Aware AI Model Optimization and Benchmarking
-
Arm Ethos-U Support in TVM ML Framework for IoT Devices
-
Powering Innovation in AI Devices on the Edge with Micro-NPUs - tinyML Asia 2020
-
TILE-MPQ: Design Space Exploration of Tightly Integrated Layer-Wise Mixed-Precision Quantized Units for TinyML Inference
