Improved Information Theoretic Generalization Bounds for Distributed and Federated Learning
Document Type
Conference Proceeding
Publication Date
1-1-2022
Abstract
We consider information-theoretic bounds on expected generalization error for statistical learning problems in a networked setting. In this setting, there are K nodes, each with its own independent dataset, and the models from each node have to be aggregated into a final centralized model. We consider both simple averaging of the models as well as more complicated multi-round algorithms. We give upper bounds on the expected generalization error for a variety of problems, such as those with Bregman divergence or Lipschitz continuous losses, that demonstrate an improved dependence of 1/K on the number of nodes. These "per node"bounds are in terms of the mutual information between the training dataset and the trained weights at each node, and are therefore useful in describing the generalization properties inherent to having communication or privacy constraints at each node.
Identifier
85132274432 (Scopus)
ISBN
[9781665421591]
Publication Title
IEEE International Symposium on Information Theory Proceedings
External Full Text Location
https://doi.org/10.1109/ISIT50566.2022.9834700
ISSN
21578095
First Page
1465
Last Page
1470
Volume
2022-June
Grant
CCF-1908308
Fund Ref
National Science Foundation
Recommended Citation
Barnes, L. P.; Dytso, A.; and Poor, H. V., "Improved Information Theoretic Generalization Bounds for Distributed and Federated Learning" (2022). Faculty Publications. 3433.
https://digitalcommons.njit.edu/fac_pubs/3433