EON Tuner - AutoML for Constrained Devices in TinyML

Explore the challenges and solutions of finding optimal machine learning models for constrained devices in this tinyML Summit 2022 talk. Dive into the EON Tuner, an AutoML tool designed specifically for tinyML applications. Learn how it searches through thousands of combinations of pre-processing, DSP, ML, and post-processing blocks to find the best model within device constraints. Discover the importance of signal processing in tinyML, the process of customizing search spaces, and evaluating models against real-world data. Gain insights into dataset quality, the balance between DSP and neural networks, and practical applications like bird sound classification and object detection for microcontrollers.

Intro
Hi, I'm Jan!
Signal processing + ML
Leveraging signal processing
ML Sensor pipeline is often combination
Wide range of parameters
Constrained targets - what's worth it?
Introducing the EON Tuner!
Understand the problem
Dataset quality
Custom search space
Bird sound classifier
Best models...
Lots of DSP? Lots of NN?
Getting started
FOMO - Object detection for MCUS
Questions?