self-taught-vit
I trained a 2.7M parameter ViT on CIFAR-10 with no labels. Just masked autoencoding. It learned to classify anyway.
- Published on
Self-supervised-learning
By tags
- felsefe (8)
- düşünme (1)
- olasılık (1)
- bilim (1)
- machine-learning (1)
- vision-transformer (1)
self-supervised-learning (1)
- mae (1)