Why are augmentations so important in contrastive SSL?

medium

Answer

$Augmentations define the invariances the encoder learns — 'same-object under transformation' becomes the training signal.$
$Strong compositions (crop + color jitter + Gaussian blur) work best (Chen et al. 2020).$
$Too weak \to trivial invariances; too strong \to destroys semantic content.$
$Rule: task-relevant invariances.$
$Cross-modal alignment (CLIP) uses text as 'super-augmentation'.$

Check yourself — multiple choice

Don't matter
Augmentations define invariances the encoder learns; strong compositions (crop + color + blur) work best; too weak/strong break representations; text = 'super-augmentation' in CLIP
Random
Only rotation

Augmentations: define invariances learned; strong compositions in SimCLR.

#representation-learning

214 interview questions in this topic.