Realising the power of edge intelligence: addressing the challenges in AI and tinyML applications for edge computing

Gibbs, M ORCID logoORCID: https://orcid.org/0009-0009-4754-7242 and Kanjo, E ORCID logoORCID: https://orcid.org/0000-0002-1720-0661, 2023. Realising the power of edge intelligence: addressing the challenges in AI and tinyML applications for edge computing. Nottingham: Nottingham Trent University. (Unpublished)

[thumbnail of 1795501_Kanjo.pdf]
Preview
Text
1795501_Kanjo.pdf - Published version

Download (1MB) | Preview

Abstract

Over the past few years, embedded systems and machine learning communities have come together to make AI ubiquitous and available near the data source, unlocking many untapped application areas that await development. As a result, hardware, software, and research have changed extremely rapidly with many recent releases of ML-enabled microcontrollers. Consequently, many frameworks have been developed for different platforms to facilitate the deployment of ML models and standardise the process. With the Artificial Intelligence of Things (AIoT) expected to grow exponentially over the next few years, more researchers and companies are expected to enter the research space. Although certain challenges of tinyML deployment can be overlooked, which makes entering the field challenging. For tinyML applications to flourish, it is important to consider how to solve entry-level challenges. The challenges below were experienced when deploying simple deep learning models to a variety of microcontrollers. These include, but are not limited to Syntiant TinyML Development Board, Sony Spresense Main Board, and Raspberry Pi Pico4ML. This poster reveals the often-overlooked challenges of tinyML and emphasizes the importance of raising awareness within the community. By addressing these issues, not only will the existing tinyML community benefit, but it will also attract a broader range of people. This will accelerate research in the field, pushing the boundaries of edge AI further and faster than ever before.

Item Type: Working paper
Creators: Gibbs, M. and Kanjo, E.
Publisher: Nottingham Trent University
Place of Publication: Nottingham
Date: 26 June 2023
Identifiers:
Number
Type
1795501
Other
Divisions: Schools > School of Science and Technology
Record created by: Jonathan Gallacher
Date Added: 29 Aug 2023 12:22
Last Modified: 07 Mar 2024 09:48
URI: https://irep.ntu.ac.uk/id/eprint/49620

Actions (login required)

Edit View Edit View

Statistics

Views

Views per month over past year

Downloads

Downloads per month over past year