Federated Learning Improves Generalization in Surgical Video Classification

The FedSurg Challenge results demonstrate that temporal modeling is the primary architectural driver for model generalization across unseen clinical centers in laparoscopic appendectomy videos. The study evaluates Federated Learning against centralized and Swarm Learning baselines using multi-center datasets.