ELL: The Microsoft Embedded Learning Library - Principles and Applications

Explore the Microsoft Embedded Learning Library (ELL) in this tinyML Summit 2019 presentation by Byron Changuion, Principal Engineering Manager of the Machine Learning and Optimization Group at Microsoft Research AI. Delve into the intricacies of AI compilers versus AI runtimes, and learn about evaluation techniques and architecture search. Discover lossless and lossy acceleration methods, including various compression techniques. Gain insights into quantization semantics, representation, and performance, with examples demonstrating the impact on weight and activation accuracy. Understand the current focus areas of ELL and its potential applications in embedded machine learning systems.

Intro
The Embedded Learning Library
Al compiler vs. Al runtime
Evaluation
Architecture search
Lossless acceleration
Lossy Acceleration mix and match compression techniques
Quantization semantics
Quantization representation
Quantization example
Quantization performance
Quantized weight accuracy
Quantized activation accuracy
Current focus areas