VICReg / Barlow Twins — non-contrastive SSL via covariance regularization.

hard

Answer

$Barlow Twins: cross-correlation of two augmented view embeddings should be identity (diagonal = 1, off-diag = 0 \to invariance + de-correlation).$
$VICReg extends with three terms: invariance (MSE between views), variance (each dim has SD > threshold), covariance (off-diag = 0).$
$No negatives, no momentum encoder — just clean regularization.$
$Elegant, competitive with contrastive methods.$

Check yourself — multiple choice

Same as SimCLR
Cross-correlation of augmented views = identity (invariance + de-correlation); VICReg adds explicit variance + covariance terms; no negatives / no momentum needed
Random
Not real

Barlow / VICReg: invariance + variance + covariance regularization; no negatives.

#representation-learning

214 interview questions in this topic.