/NeuralPCA

Neural PCA accepted to IJCAI 2022

Neural-PCA

This repository contains the PyTorch implementation of Neural-PCA proposed in the IJCAI2022 paper:

Shen Li, Bryan Hooi. Neural PCA for Flow-Based Representation Learning. IJCAI2022

Empirical Results

Neural-PCA provides better architectural inductive bias for normalizing flow, which leads to better data representation without compromising generative performance.

Requirements

  • python==3.6.0
  • torch==1.6.0
  • torchvision==0.7.0
  • tensorboard==2.4.0

Getting Started

Training

Citation

@article{li2022neural,
  title={Neural PCA for Flow-Based Representation Learning},
  author={Li, Shen and Hooi, Bryan},
  journal={IJCAI 2022},
  year={2022}
}