EasyDeepLearn

SimCLR — recipe.

medium

Answer

  • Contrastive framework for visual SSL: (1) two random augmentations of each image → positive pair, (2) encoder + projection head, (3) NT-Xent loss (InfoNCE variant): pull positives together, push all other batch items apart.
  • Needs LARGE batch size (~4096) for enough negatives.
  • Removes projection head at downstream time.
  • Started the modern SSL for vision revolution; superseded by SimSiam / DINO / MAE.
Check yourself — multiple choice
  • Random
  • Two augmentations → positive pair; encoder + projection head + NT-Xent loss; large batch (~4096) for negatives; started modern vision SSL era
  • Same as classifier
  • Not real

SimCLR: aug-pair contrastive + projection head + big batch (~4096) NT-Xent.

#representation-learning

Practise Unsupervised Learning

214 interview questions in this topic.

Related questions