This is an un-official implementaion of the i-MIX : A DOMAIN -A GNOSTIC STRATEGY FOR CONTRASTIVE REPRESENTATION LEARNING-using pytorch.
I got accuaracy of 72% on Tabular data as the one in the paper with SimCLR . More experiments coming soon.
Major modules implemented in the code
- SimCLR Loss
- Tabular Model
- Augementation for tabuler data
- Mixup
- Mixup with SimCLR loss
- Pre-training & Fine-tuning
- Use other modalities (images , speech)
- Experiment Moco Loss and Byol Loss
- Experiment different data sizes
- Experiment with imagea without augmentation
- covert code to .py
- Improve documentation