Obaidullah Zaland, Zulfiqar Ahmad Khan, Monowar Bhuyan
This paper introduces One-Shot Incremental Federated Learning (OSI-FL), a framework that addresses communication overhead and catastrophic forgetting in federated learning by using category-specific embeddings and selective sample retention.
In today's data-driven world, managing large and diverse data streams while maintaining privacy is a big challenge. Federated learning is a popular approach because it allows for collaborative model training without sharing raw data. However, it struggles with incremental data and communication limits. This research presents a new method called One-Shot Incremental Federated Learning (OSI-FL), which reduces communication needs and prevents the model from forgetting previously learned tasks. It uses specialized data embeddings and a technique to keep the most useful data samples for training. This approach has shown better performance than existing methods in various scenarios.
