Information Content Maximization

Definition

A principle for preventing collapse in self-supervised learning where the objective is to maximize the information content of learned representations. Can be done dimension-wise (making variables independent/disentangled) or sample-wise (making samples different). Information content itself is difficult to measure (only upper bounds available) but approximations work well.

Cast within

No topic-region cast yet — this appears once Information Content Maximization's compiled claims are aligned into a topic region's argument tree.

Claims mentioning Information Content Maximization (2)

Your notes

Loading notes...