This is the official repository accompanying the NeurIPS 2020 Paper Functional Regularization for Representation Learning: A Unified Theoretical Perspective.
This directory contains:
- The scripts to generate synthetic data with the properties described in Section 5.1 of the paper
- The scripts to run end-to-end training and training with functional regularization via an auto-encoder
- The scripts to plot the t-SNE visualization graphs for the functional approximations
This directory contains:
- The scripts to generate synthetic data with the properties described in Section 5.2 of the paper
- The scripts to run end-to-end training and training with functional regularization via masking the first input component
- The scripts to plot the t-SNE visualization graphs for the functional approximations
Work-in-progress.
Use the transformers code repository released by HuggingFace.
The MRPC dataset can be accessed here.
If you find this code or our paper useful, please consider citing us using:
@article{garg20functional,
author = {Siddhant Garg and Yingyu Liang},
title = {Functional Regularization for Representation Learning: A Unified Theoretical Perspective},
conference = {NeurIPS 2020},
url = {https://arxiv.org/abs/2008.02447},
}
For direct communication, please contact me (sidgarg is at amazon dot com) if you have any questions regarding the code or the experiments.