DeepMom is a pipeline for robust deep learning. It is described here: DeepMoM: Robust Deep Learning With Median-of-Means. This repository provides the corresponding implementations.
The code in SimulationsRegression.R provides a comparison of least-squares, Huber, and least-absolute deviation estimators to our ReLU-based DeepMoM estimators in regression problems;
the code in SimulationsClassification.R provides a comparison of soft-max cross-entropy estimators to our ReLU-based DeepMoM estimators in classification problems. (The simulation can take a while to complete on a single machine.)
The code in TcgaApplication.R applies DeepMoM to seven TCGA data sets. (Takes a while to complete if on a single machine.)
AdditionalFunctions: The source code of the functions required for computing DeepMoM.
TcgaData: A tutorial R markdown file for downloading TCGA datasets.
-
Shih-Ting Huang, Ph.D. student of Mathematical Statistics, Ruhr-University Bochum
-
Johannes Lederer, Professor of Mathematical Statistics, Ruhr-University Bochum
The code in this repository is written in R with version R 4.1.2 and supports all plarforms which are supported by R itself.
This repository does not depend on any R libraries or external sources.
Some of our results are based upon data generated by the TCGA Research Network: https://www.cancer.gov/tcga.
All codes are licensed under the MIT license. To
view the MIT license please consult LICENSE.txt.
The paper can be found here: DeepMoM: Robust Deep Learning With Median-of-Means
It should be cited as "Huang, S.-T. and Lederer, J., 2021. DeepMoM: Robust Deep Learning With Median-of-Means. arXiv:2105.14035."