Overview
Explore the cutting-edge world of tinyML in this 35-minute conference talk from the tinyML Summit 2022. Delve into the concept of creating ultra-compact machine learning models, specifically those not exceeding 1 kB in size. Learn how Neuton, a no-code platform, revolutionizes the development of tiny smart devices by automating the creation of optimal ML models. Discover the process of embedding these compact models into memory-constrained hardware, even with 8 and 16-bit precision, without the need for compression, quantization, or pruning. Through a practical demonstration focused on food quality determination, gain insights into the end-to-end process of developing and implementing super tiny ML models in 8-bit sensor microcontrollers. Understand the historical context of tinyML, its future positioning, and how to leverage existing frameworks for enhanced results. This talk, presented by Blair Newman, CTO of Neuton, offers valuable knowledge for embedded engineers and those interested in the expanding capabilities of tiny smart devices across various domains.
Syllabus
Intro
Welcome
TinyML History
How to position TinyML for the future
Where are you on your TinyML journey
What does embedding a model mean
The ideal weight for a tinyML model
Size is not enough
What is Newton
Benefits of Newton
Opportunities
Leveraging existing frameworks
Results
How do we do this
No coding required
Use case
Summary
Sponsors
Taught by
tinyML