A vector-contraction inequality for Rademacher complexities
read the original abstract
The contraction inequality for Rademacher averages is extended to Lipschitz functions with vector-valued domains, and it is also shown that in the bounding expression the Rademacher variables can be replaced by arbitrary iid symmetric and sub-gaussian variables. Example applications are given for multi-category learning, K-means clustering and learning-to-learn.
This paper has not been read by Pith yet.
Forward citations
Cited by 5 Pith papers
-
Sample Complexity of Scientific Discovery: PAC Learnability of Compositional Function Trees
Proves that Rademacher complexity of depth-d compositional trees over finite operator vocabulary is controlled by (K b L)^{d} / sqrt(n) under Lipschitz conditions on operators.
-
Symmetrization of Loss Functions for Robust Training of Neural Networks in the Presence of Noisy Labels
Symmetrization of multi-class losses produces a unique convex symmetric loss that locally approximates others and supports robust neural training under label noise.
-
Smoothness-Based Derandomization of PAC-Bayes Bounds
Derives smoothness-based PAC-Bayes bounds for deterministic predictors by bounding the Jensen gap class via Rademacher complexity, yielding flatness terms in Jacobians/Hessians, and proposes a corresponding regularize...
-
Smoothness-Based Derandomization of PAC-Bayes Bounds
Derives smoothness-based PAC-Bayes derandomization bounds for deterministic predictors using Rademacher complexity of the Jensen gap class, yielding Jacobian/Hessian flatness terms and a practical regularizer tested o...
-
Symmetrization of Loss Functions for Robust Training of Neural Networks in the Presence of Noisy Labels
Symmetrizing cross-entropy produces the unique convex multi-class unhinged loss, which locally approximates other symmetric losses, and enables new interpolating losses SGCE and alpha-MAE with competitive performance ...
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.