SciTinyML-22: Scientific Use of Machine Learning on Low-Power Devices

A TinyMLedu Workshop

Spring/Summer 2022

SciTinyML-22: Scientific Use of Machine Learning on Low-Power Devices

A TinyMLedu Workshop

Spring/Summer 2022

Course Overview

SciTinyML-22 is an ICTP Virtual Meeting supported by the TinyML4D Academic Network and open to all exploring how embedded ML (tinyML) can impact the developing world through hands-on activities using embedded hardware devices. SciTinyML-22 ran regionally in 2022 with seperate workshops for Africa (187 participants from 29 countries), Asia, and Latin America. This program is a collaboration led by the Abdus Salam International Centre for Theoretical Physics (ICTP), the Harvard John A. Paulson School of Engineering and Applied Sciences, and TinyMLedu.

TinyML is a subfield of Machine Learning focused on developing models that can be executed on small, realtime, low-power, and low-cost embedded devices. This allows for new scientific applications to be developed at an extremely low cost and at large scale. The TinyML process starts with collecting data from IoT devices, then training the collected dataset to extract knowledge patterns; these patterns are then packaged into a TinyML model that considers the target microprocessor’s limited resources such as memory and processing power. The resulting model is then deployed on embedded devices where it is used to evaluate new sensor data in real-time. Typically, power requirements are in the mW range and below which enables a variety of use-cases targeting battery operated devices. TinyML represents a collaborative effort between the embedded power systems and Machine Learning communities, which traditionally have operated independently.

Workshop topics include an intorduction to (tiny)ML concepts, getting started with the TinyML training kit, examples of TinyML applications, and scientific applications of ML.

My Roles