Code for Supercharging Imbalanced Data Learning With Energy-basedContrastive Representation Transfer (NeurIPS 2021)

The algorithm is built with:

• Python (version 3.7 or higher)
• Numpy (version 1.16 or higher)
• PyTorch (version 1.3.1)

Clone the repository, e.g.:

git clone https://github.com/ZidiXiu/ECRT.git

Here we present a toy synthetic dataset

python train train_CRT_toy.py

When building the CRT framework, we referenced the following sources:

• MADE flow
• GCL learning