Interpolation can hurt robust generalization even when there is no noise

Authors: Konstantin Donhauser, Alexandru Tifrea, Michael Aerni, Reinhard Heckel, Fanny Yang

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

Reproducibility Variable Result LLM Response
Research Type Experimental We plot the robust accuracy gain of (a) early-stopped neural networks compared to models at convergence, fit on sanitized (binary 1-3) MNIST that arguably has minimal noise; and ℓ2 regularized estimators compared to interpolators with λ 0 for (b) linear regression with n = 103 and (c) robust logistic regression with n = 103.
Researcher Affiliation Academia 1ETH Zurich 2Rice University 3Technical University of Munich
Pseudocode No The paper does not contain any pseudocode or algorithm blocks.
Open Source Code No The paper does not provide an explicit statement or link for the open-sourcing of the described methodology's code.
Open Datasets Yes sanitized (binary 1-3) MNIST
Dataset Splits No The paper does not explicitly provide specific training/validation/test dataset splits (percentages, sample counts, or references to predefined splits) in the main text.
Hardware Specification No The paper does not provide specific hardware details (e.g., exact GPU/CPU models, processor types, or memory amounts) used for running its experiments in the main text.
Software Dependencies No The paper does not provide specific software dependencies with version numbers (e.g., library or solver names with versions) needed to replicate the experiment.
Experiment Setup No The paper mentions that experimental details are in Appendix B, which is not provided. The main text describes some model and data parameters (e.g., ϵ values, d/n ratios) but does not provide specific training hyperparameters such as learning rate, batch size, or number of epochs.