Concept
Positive Semi-Definite Directional Regularizer
A concept in optimization theory that adds a positive semi-definite matrix-weighted penalty term to the loss function, imposing selective constraints on the directions of parameters or representations, with stronger penalties for directions that are less reconstructable.
Timeline (last 90 days)
Nov 2
When linear reconstruction is used and collaborators' representations are fixed, CL's loss function is equivalent to a positive semi-definite directional regularizer that imposes stronger constraints on directions with lower reconstructability.
Unverified50%