Distributed Learningedit
Learning with data, computation, or evaluation distributed across participants.
Distributed Learning covers learning settings in which data, computation, or optimization steps are spread across multiple clients, devices, institutions, or workers. It includes decentralized and federated learning, which differ in their coordination and data-access arrangements.1
Research contextedit
Distributed learning is a foundation of AI and networks. Networked AI has different constraints from centralized training: communication can be expensive, local data can be non-identically distributed, and privacy or ownership may limit what can be shared. These constraints make data selection and pruning more important, because transmitting or training on all available data may be impractical.
Connection to Qiao's workedit
This perspective appears in Qiao's synthetic-data work, where multiple silos must reason about distributional drift without a pooled dataset. Distributed learning therefore provides the infrastructure context for Qiao's current AI and networks direction.
See alsoedit
Footnotesedit
-
McMahan et al., "Communication-Efficient Learning of Deep Networks from Decentralized Data", AISTATS 2017, is a standard reference point for federated learning and communication-efficient decentralized optimization. ↩