AI RESEARCH
Objective-Specific Privileged Bases via Full-Prefix Matryoshka Learning
arXiv CS.LG
•
ArXi:2605.09160v1 Announce Type: new Learned representations are often invariant to rotational transformations, leaving individual dimensions non-identifiable and interchangeable. We study how Matryoshka Representation Learning (MRL) induces a task-aligned privileged basis distinct from variance-based or regularizer-induced orderings. In the linear setting, we prove that full-prefix MRL recovers the ordered principal directions, and can be computed efficiently using shared statistics.