Federated Active Learning in Multitask Environments

With the constantly growing need for knowledge-sharing without data-sharing, Federated Learning is becoming more and more important. The need for Federated methods in Active Learning arises naturally from multi-agent scenarios, where the agents need common and local knowledge simultaneously. Both areas are closely related to Multitask learning because different agents can always have somewhat different goals. Despite their relatedness and close relation to budgeted learning and game theory, the synthesis of these areas is still unsolved. In this research, the student will explore generalizations of Multitask Federated Learning, where partners actively can request further samples based on their private preferences and cost models, performing a collaborative exploration.