A Universal Law of Robustness via Isoperimetry

Authors: Sebastien Bubeck, Mark Sellke

NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Theoretical We propose a theoretical explanation for this phenomenon. We prove that for a broad class of data distributions and model classes, overparametrization is necessary if one wants to interpolate the data smoothly. Namely we show that smooth interpolation requires d times more parameters than mere interpolation, where d is the ambient data dimension. We prove this universal law of robustness for any smoothly parametrized function class with polynomial size weights, and any covariate distribution verifying isoperimetry (or a mixture thereof).
Researcher Affiliation Collaboration S ebastien Bubeck Microsoft Research sebubeck@microsoft.com Mark Sellke Stanford University msellke@stanford.edu
Pseudocode No The paper does not include any pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any statements about releasing code or links to a code repository for the methodology described.
Open Datasets Yes To put Theorem 1 in context, we compare to the empirical results presented in [MMS+18]. In the latter work, they consider the MNIST dataset which consists of n = 6 104 images in dimension 282 = 784.
Dataset Splits No This paper is theoretical and focuses on proving a mathematical law. It discusses existing empirical results from other papers (e.g., [MMS+18]) but does not define or use its own dataset splits (train/validation/test) to reproduce experiments.
Hardware Specification No The paper is theoretical and does not conduct experiments, therefore no hardware specifications are mentioned.
Software Dependencies No The paper is theoretical and does not describe any experimental implementation, thus no software dependencies with version numbers are listed.
Experiment Setup No The paper is theoretical and does not describe specific experiments with hyperparameters or training configurations conducted by the authors.