Local Rademacher complexities
Peter L. Bartlett, Olivier Bousquet, Shahar Mendelson
Abstract
We propose new bounds on the error of learning algorithms in terms of a data-dependent notion of complexity. The estimates we establish give optimal rates and are based on a local and empirical version of Rademacher averages, in the sense that the Rademacher averages are computed from the data, on a subset of functions with small empirical error. We present some applications to classification and prediction with convex function classes, and with kernel classes in particular.
Create a lesson
Related papers
Sequential Certification of Partially Identified Product Targets: Information Limits and Adaptive Sampling
Abdallah Skiredj, Abderrahman Skiredj, Amal El Fallah Seghrouchni
A Partial-Matching Proof of the Neyman--Pearson Lemma
Rami V. Tabri
Neyman Orthogonalized Specification Tests for Conditional Tail Index Models
Liuhua Peng, Shuo Li, Xiaojun Song
On Natural Exponential Families with Log-Affine Generalized Variance
Zeyu Cao, Shaosai Huang
Fundamental Limits of Coherent Imaging in The Presence of Atmospheric Turbulence
Songbin Liu, Hanyu Yang, Junren Chen et al.
Regularization vs. Convolution in Kernel-Based Hypothesis Testing
Omar Hagrass, Bharath K. Sriperumbudur