What is it about?
Energy Harvesting device running on different harvesting power need to adapt the AI Algorithm with the varying environment. For example, in low harvesting power (rainy weather) a light-weight AI model needs to execute to maintain the quality of service.
Featured Image
Photo by Biel Morro on Unsplash
Why is it important?
1. Enable AI in EH device is difficult due to resource limitation. 2. Traditional way of environment adaptive AI is more difficult because it will require multiple AI model to deploy in same EH resource. 3. Therefore, we propose an idea of weight sharing between AI models to lower the memory footprint. 4. We also show how to design and implement shared-weight models.
Perspectives
This is a software/hardware co-design project. There are a lot of work going on to reduce the size of the AI models. I think this project will give an insight to develop scalable/adaptable AI for the small devices while maintaining the accuracy and resource constraints.
Sahidul Islam
University of Texas at San Antonio
Read the Original
This page is a summary of: EVE, October 2022, ACM (Association for Computing Machinery),
DOI: 10.1145/3508352.3549451.
You can read the full text:
Resources
Contributors
The following have contributed to this page
