Assisted Learning: A Framework for Multiple Organizations Learning

Large-scale multimodal data are being rapidly generated from the interaction between humans and machines. Sharing heterogeneous data among multiple organizations for collaborative learning typically faces tradeoffs between learning efficiency and data privacy. In general, for a method to achieve the optimal performance on one aspect will sacrifice another one. To tackle this challenge, we introduce the Assisted Learning framework where a service provider Bob assists a user Alice with supervised learning tasks without transmitting Bob's private algorithm or data. Bob assists Alice either by building a predictive model using Alice's labels or by improving Alice's learning through iterative transmissions of task-specific statistics. Theoretical analysis shows that the proposed method can achieve lossless learning performance for certain models. We also demonstrate the wide applicability of the proposed approach by various experiments, including real-world medical benchmarks.