What is triplet loss and its main challenge?

hard

Answer

$L(a, p, n) = max(d(a, p) - d(a, n) + margin, 0), where a=anchor, p=positive, n=negative.$
$Wants the anchor closer to positives than to negatives by at least the margin.$
$Main challenge: 'triplet mining' — most random triplets are easy (already satisfied), so you must actively sample semi-hard / hard triplets to keep training progressing.$
$Standard in face recognition (FaceNet).$
$Superseded in most tasks by info-NCE / softmax-based losses.$

Check yourself — multiple choice

Uses two examples
Anchor + positive + negative; hard-triplet mining is the practical challenge; used in face recognition
Only two negatives
No margin needed

Triplet: d(a,p) < d(a,n) - margin; hard mining is key.

#losses#contrastive

214 interview questions in this topic.