35 min
Seminar, Hosted by Simons Institute for the Theory of Computing
The Many Faces of Heterogeneity: Federated, Continual, and Modular Learning
Vector Institute
Recording
Abstract
Marco Ciccone connects client drift in federated learning, forgetting across time, and unreliable composition of independently trained modules. These settings each require a system to learn from heterogeneous sources while preserving or combining useful knowledge. The talk shows where methods from one setting can inform the others and outlines questions for designing around heterogeneity.