molierflower/np-match

A Pytorch implementation of ICML 2022 paper "NP-Match: When Neural Processes meet Semi-Supervised Learning"

NP-Match

NP-Match: When Neural Processes meet Semi-Supervised Learning

Jianfeng Wang¹, Thomas Lukasiewicz¹, Daniela Massiceti², Xiaolin Hu³, Vladimir Pavlovic⁴ and Alexandros Neophytou⁵

University of Oxford¹, Microsoft Research², Tsinghua University³, Rutgers University⁴, Microsoft⁵

In ICML 2022

Arxiv

Build

please run with the following command:

conda env create -f NP-Match.yml
conda activate NP-Match

Experiment

To run experiments on small datasets, e.g., CIFAR, STL-10, please run with their corresponding config files, stored in "config/npmatch/".

For example:

python npmatch.py --c config/npmatch/npmatch_cifar10_250_0.yaml

To run experiment on ImageNet, please prepare the dataset, and change the config file accordingly. Then, run with

python npmatch.py --c config/npmatch/npmatch_imagenet_100000.yaml

For the pertrained models, they were not saved. The pretrained models will be relased later once we have enough resources for training.

Citation

@inproceedings{wang2022np,
title={NP-Match: When Neural Processes meet Semi-Supervised Learning},
author={Wang, Jianfeng and Lukasiewicz, Thomas and Massiceti, Daniela and Hu, Xiaolin and Pavlovic, Vladimir and Neophytou, Alexandros},
booktitle={International Conference on Machine Learning},
pages={22919--22934},
year={2022},
organization={PMLR}
}