From the Cloud to the Edge: Reverse-Engineering and Downsizing a Black Box ML Algorithm for Sleep Stage Prediction

Watch a conference talk exploring the fascinating journey of reverse-engineering and miniaturizing a proprietary sleep stage prediction algorithm. Learn how Edge Impulse tackled the challenge of improving and scaling down an existing black box ML model without access to original data or labels, while working within the constraints of clinically validated data. Discover practical strategies for cost-effective data collection, transitioning computation from cloud to edge devices, and implementing new features like wake-up on light sleep. Follow along as Jan Jongboom, Co-founder and CTO of Edge Impulse, demonstrates leveraging powerful models like YOLOV5, Whisper, and Segment Anything for labeling, setting up robust pipelines, implementing feature extraction, and optimizing the development feedback loop to enhance existing market devices with edge AI capabilities.