AIM of the Project

• To implement Representation Learning algorithms (Contrastive and self-supervised learning algorithms)

Algos Implemented

• SimCLR (ICML 2020)
• SupCon (NIPS 2020)
• SimSiam (CVPR 2021)
• Bootstrap your own Latent (BYOL) (NIPS 2020)
• Barlow Twins (ICML 2021)
• TripletMarginLoss (CVPR 2015)

Results

Algorithm	CIFAR10 (R50)	CIFAR100 (R50)	CIFAR10 (R18)	CIFAR100 (R18)
SimCLR	87.5	57.7	85.9	55.0
SupCon	94.0	74.7	93.5	70.4
Triplet	83.4	76.3	86.0	64.5
Barlow Twins	81.2	47.7	80.3	45.8
BYOL	83.0	47.0	84.8	54.8
SimSiam	76.5	34.5	88.6	62.3
DARe (Ours)	89.4	62.3	87.3	61.6

Extension

• Distribution Alignment Regularizer (DARe) contrastive learning

Team Members

• Mansi Tomer (SR: 24014)
• Nirbhay Sharma (SR: 24806)