Case Study: Scaleout cuts model updates from weeks to hours with Ultralytics YOLO

Scaleout cuts model updates from weeks to hours with Ultralytics

Scaleout, a company building infrastructure for distributed machine learning, faced the challenge of updating AI models in environments where sensitive data, such as defense or industrial footage, could not be moved to a central server due to privacy, security, or bandwidth constraints. Their challenge was to continuously improve detection models without ever moving the raw data from the edge devices. They turned to Ultralytics YOLO models to solve this.

The solution implemented by Ultralytics involved using their lightweight YOLO models, specifically YOLOv8 nano, within Scaleout's federated learning platform. This allowed models to be fine-tuned directly on edge hardware at the data source. The result was a dramatic reduction in model update size to just ~10 MB and a turnaround time slashed from weeks to mere hours, enabling rapid iteration and improved detection without the raw data ever leaving the secure site.

Scaleout