ReLU Regression with Massart Noise

Authors: Ilias Diakonikolas, Jong Ho Park, Christos Tzamos

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

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate that our algorithm significantly outperforms naive applications of 1 and 2 regression on both synthetic and real data. In this section, we apply our algorithms that are based on radial-isotropic transformations to both synthetic and real datasets and compare robustness in regression with other baseline methods of 1 and 2-regression. Our experiments demonstrate the efficacy of radial-isotropic transformations in robust regression and how our algorithms outperform baseline regression methods.
Researcher Affiliation Academia Ilias Diakonikolas University of Wisconsin-Madison ilias@cs.wisc.edu Jongho Park University of Wisconsin-Madison jongho.park@wisc.edu Christos Tzamos University of Wisconsin-Madison tzamos@wisc.edu
Pseudocode Yes Algorithm 1 Linear function recovery via radial isotropy
Open Source Code No The paper does not contain any explicit statement about making the source code available or provide a link to a code repository.
Open Datasets Yes The drug discovery dataset was originally curated by (40) and we used the same dataset as the one used in (14).
Dataset Splits Yes The dataset has a training and test set of 3084 and 1000 points of 410 dimensions.
Hardware Specification Yes All experiments were done on a laptop computer with a 2.3 GHz Dual-Core Intel Core i5 CPU and 8 GB of RAM.
Software Dependencies No The paper mentions 'CVXPY s linear program solver' but does not specify a version number or other software dependencies with versions.
Experiment Setup Yes The experiment is set up with = 0.4, w = 9e2 + Pd i=1 ei, w0 = 0, 240 samples from Dx, and gradient descent step size of one. For Original , we use a step size of 1/465 to keep the magnitude of the points xi comparable to that of the transformed points xi.