https://github.com/Sherrylone/m-mix

SIGKDD 2022 paper
https://github.com/Sherrylone/m-mix

Last synced: 3 months ago
JSON representation

SIGKDD 2022 paper

• Host: GitHub
• URL: https://github.com/Sherrylone/m-mix
• Owner: Sherrylone
• Created: 2022-05-26T01:59:32.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2022-09-29T07:52:27.000Z (about 2 years ago)
• Last Synced: 2024-05-14T00:16:06.032Z (6 months ago)
• Language: Python
• Homepage:
• Size: 1.17 MB
• Stars: 9
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• Awesome-Mixup - [Code

README

        
# M-Mix: Generating Hard Negatives via Multi-sample Mixing for Contrastive Learning

Code of SIGKDD 22 paper "[M-Mix: Generating Hard Negatives via Multi-sample Mixing for Contrastive Learning](https://sherrylone.github.io/assets/KDD22_M-Mix.pdf)"

![M-Mix](./framework.png)

This paper proposes to mix multiple samples in one mini-batch to generate hard negative pairs.

To pre-train the encoder on CIFAR-10 and CIFAR-100, run:

```

python main.py --dataset cifar10 (cifar100) --threshold 0.9

```

The config `--threshold 0.9` is used for selecting negative samples to mix.

For graph and node classification. Run: 

```

python main.py

```

You should download the dataset by yourself.