EasyDeepLearn

What is consistency regularization?

medium

Answer

  • Enforce that two augmented views of the same unlabeled input produce similar predictions: Lcons  =  DL_{\mathrm{cons}}\; = \;D(f(aug1(x)), f(aug2(x))) with D = KL or MSE.
  • Powers modern semi-supervised methods (FixMatch, Mean Teacher, MixMatch) and self-supervised learning (SimCLR, MoCo, BYOL).
  • Cornerstone technique when labeled data is scarce.
Check yourself — multiple choice
  • Consistency means identical labels
  • Two augmentations of the same input should predict similarly — cornerstone of SSL / semi-supervised
  • Only for supervised learning
  • Adds noise to labels

Consistency: two views ⇒ similar predictions — key to FixMatch / SimCLR / BYOL.

#regularization#self-supervised#augmentation

Practise Deep Learning

214 interview questions in this topic.

Related questions