Self-Supervised & Representation Learning

Contrastive learning (SimCLR, InfoNCE) and non-contrastive methods (VICReg).

0 / 4 solved

Question Tags Difficulty Success

1. One Layer Upstream Of The Loss

simclrcontrastive-learning +1

Hard —

2. Which VICReg term directly prevents complete collapse?

Machine LearningRepresentation Learning

Medium —

3. Why can an InfoNCE mutual-information bound saturate?

Machine LearningRepresentation Learning

Hard —

4. Stop, Collaborate, Don't Collapse

self-supervised-learningcontrastive-learning +1

Hard —

Learning Resources

CS231n, self-supervised learning

Up next

4 of 4 questions here are still unsolved.

Next module Causal Inference