How do you evaluate the quality of a self-supervised representation?

medium

Answer

$(1) Linear probing: freeze encoder, train linear classifier on downstream labels \to measures 'linearly separable info'.$
$(2) Fine-tuning: unfreeze, full downstream training \to measures 'total info'.$
$(3) k-NN classification in embedding space \to non-parametric.$
$(4) Transfer to many tasks (VTAB, ELEVATER).$
$(5) Robustness under distribution shift.$
$Rule: report linear probe + k-NN + fine-tune; don't over-index on one downstream task.$

Check yourself — multiple choice

Just accuracy
Linear probe (linearly-separable info) + fine-tune (total info) + k-NN (non-parametric) + transfer to VTAB / ELEVATER + robustness; report multiple, not one task
Random
Only silhouette

SSL evaluation: linear probe + k-NN + fine-tune + transfer benchmarks.

#representation-learning#evaluation

Practise Unsupervised Learning

214 interview questions in this topic.

Read the lesson Flashcards Take the quiz

Related questions