Skip to content
ml-recall

#representation-geometry

1 question from Transfer Learning

Timed test

1 question · 1 min · score and explanations at the end

Take a test
Q 1 / 1Score 0 / 0

Question 1: Single choice

hardAI-checkedTransfer Learning

An encoder ff maps data from NN groups to unit vectors; group ii lives on a finite set SiS_i. Define diameteri=max⁡s,t∈Si∥f(s)−f(t)∥\mathrm{diameter}_i = \max_{s, t \in S_i}\lVert f(s) - f(t) \rVert and margini→j=min⁡s∈Si, t∈Sj∥f(s)−f(t)∥\mathrm{margin}_{i \to j} = \min_{s \in S_i,\, t \in S_j}\lVert f(s) - f(t) \rVert. A contrastive "classification" draws an anchor and a positive from group ii and one negative from every other group, and is correct if the positive has the largest dot product with the anchor. Which condition guarantees that every such classification is correct?

Choose one option