TupletMarginLoss - Githubissues

Multi-Objective-NAS / self-supervised-nas

Official implementation of the paper "Pretraining Neural Architecture Search Controllers with Locality-based Self-Supervised Learning" (NeurIPSW 2020)

5 stars 0 forks source link

TupletMarginLoss #5

Closed Naruu closed 3 years ago

Naruu commented 3 years ago

Tuplet Margin losss는 miner가 필요하지 않음(Loss function에서 알아서 가중치를 줌)

batch size대로 tuplet을 만들기 때문에 batch size가 작음

epoch, optimizer lr 늘려서 다시 실험