self-taught-vitPublished onFebruary 8, 2025I trained a 2.7M parameter ViT on CIFAR-10 with no labels. Just masked autoencoding. It learned to classify anyway.#machine-learning#vision-transformer#self-supervised-learning
