This project will investigate whether multiple NVIDIA Jetson edge devices can collaboratively train and improve an AI occupancy-detection model without transferring raw video or image data between sites. The student will build and evaluate a federated learning setup in which devices train locally and contribute model updates to a shared global model, while accounting for differences in hardware capability, network conditions, and locally available data.
The project will compare federated learning methods, measure model performance and communication cost, and examine how local training can be scheduled without disrupting real-time occupancy inference. Depending on scope, the student may also investigate privacy-preserving techniques such as secure aggregation or differential privacy.
The project would suit a student interested in federated learning, edge AI, computer vision, distributed systems, and privacy-preserving machine learning.