TinyML Auto ML Forum: Build Neural Networks Less Than 1KB

Neuton’s team, represented by our CTO, Blair Newman, and Head of Machine Learning & Analytics, Danil Zherebtsov, was super excited to kick off a series of deep-dive sessions organized by the tinyML Foundation as part of the tinyML Auto ML Forum. 

Blair gave a cool overview of our unique framework and its peculiarities, followed by a brief demo of how Neuton.AI solves a handwritten digit recognition task by Danil. 

The key ideas to catch from this session:

• More than 96% of existing cases that claim to be TinyML do not meet these requirements, which implies that most users are still unfamiliar with tools for creating models deployable on memory-constrained hardware.

 

• TinyML tasks require building tiny models without sacrificing accuracy, which means finding both the ideal size and best metric.

 

• The ideal weight of a TinyML model is <1Kb and our patented neural network framework, Neuton, empowers users to automatically produce such models in one iteration and without compression.