Explore resource-efficient machine learning techniques for IoT devices with limited RAM in this 22-minute conference talk from the tinyML Summit 2020. Delve into the compute spectrum of AI, requirements for edge computing, and broad approaches for TinyML. Learn about Microsoft's EdgeML Library and its building blocks, including ProtoNN training algorithm and FastRNN. Discover how these methods compare to uncompressed techniques in terms of prediction accuracy and model size. Gain insights into implementing ML on edge devices for time series data and understand the potential applications and conclusions drawn from this research.
Resource Efficient Machine Learning in a Few KBs of RAM - tinyML Summit 2020
Overview
Syllabus
Intro
Compute Spectrum: AI
Resource-constrained lot Devices
Requirements on The Edge
Broad approaches for TinyML
Edge Machine Learning (EdgeML) - Objectives
Microsoft's EdgeML Library
EdgeML Building Blocks
ProtoNN: Training Algorithm
Comparison to Uncompressed Methods
Prediction Accuracy vs Model Size
FastRNN
Prediction on Edge Devices
Time Series
Conclusions
Taught by
tinyML
