Project Overview
This research project explores the potential of TinyML to revolutionize educational assessment in resource-constrained environments. Our work focuses on developing offline, automated grading systems that can operate without reliable internet connectivity while maintaining privacy, cost-effectiveness, and sustainability.
The project addresses the critical need for accessible educational technology solutions that can function in diverse global contexts, from rural schools with limited infrastructure to privacy-conscious institutions seeking to keep student data local.
Literature Review completed
Our comprehensive literature review examined the current state of TinyML applications in educational and broader contexts, analyzing the hardware and software ecosystem while identifying research gaps in offline AI tools.
Research Questions
- RQ1 — What is the current state of the TinyML hardware and software ecosystem?
- RQ2 — How do existing modules compare in terms of performance, energy consumption, and cost effectiveness?
- RQ3 — What specific technical requirements and constraints must be addressed to adapt TinyML systems to educational outcomes such as automated assessment tasks?
Research Tools
- Zotero — A reference manager for collecting, organizing, annotating, citing and sharing research
- Rayyan — An collaborative online systematic literature review for paper screening and data extraction
- Overleaf — A collaborative online LaTex editor for coauthoring a paper
Publication
We have written a paper on our findings and are currently preparing it for submission to journals. As such we are unable to share our paper at this time. However, this website will be updated in the future with details on where and how to access the paper.
Proof of Concept completed
We developed a functional demonstration of automated grading using TinyML/Edge AI devices. The system captures images of student homework, converts handwriting to text using OCR, and evaluates responses against a rubric. All of this works offline, in a web browser, using modest computational resources.
System Requirements
- Internet connection: required to download the PWA and language model, after which it can be used offline.
- Operating system: Windows 10 or 11, macOS 13+, Linux, or ChromeOS 16389.0.0+. ChromeOS is only supported on Chromebook Plus devices.
- Browser: Google Chrome 142+. This is the only browser capable of running Gemini Nano models locally.
- Hardware: Any desktop, laptop, tablet, or Chromebook Plus device capable of running the required browser and operating system. GPU is not required, but the device should have at least 4GB of VRAM, 16GB of RAM, and 4 CPU core. Mobile devices are not supported.
The proof of concept is available as a Progressive Web App (PWA) that can be installed on supported devices.
Visit the PWA link to try it out or view a walkthrough.
Demo Video completed
We recorded a demo video of the proof of concept that demonstrates ingesting a sample homework image and seeing real-time OCR processing and rubric-based evaluation.