SimSiam — what makes it minimal?

hard

Answer

$Removes EMA target: both branches share weights.$
$Uses only stop-gradient on one branch and an asymmetric predictor head.$
$Still matches BYOL / SimCLR.$
$Chen & He 2021 showed non-collapse is achieved purely by stop-gradient asymmetry — huge simplification.$
$Foundation of understanding why SSL works: asymmetric optimization dynamics, not contrast, is the key.$

Check yourself — multiple choice

Same as SimCLR
No negatives + no EMA target + stop-gradient on one branch + predictor head; matches BYOL/SimCLR → stop-grad asymmetry is core mechanism
Random
Not real

SimSiam: no negatives / no EMA; stop-grad + asymmetric predictor; matches BYOL.

#representation-learning

214 interview questions in this topic.