Description

This repository contains the code for reproducing the experiments for the paper Causal Models and Adversarial Training: Selecting the right properties for robust non-topical text classification.

It contains two sections:

1. experiments with the adversarial methods,

2. experiments with CausaLM.

For CausaLM, we partly use code from the following repo: https://github.com/amirfeder/CausaLM

In this repo, we apply approaches described in the following papers:

https://arxiv.org/abs/2112.01406

https://arxiv.org/abs/2005.13407