Deeper Learning

Lecture 4 Transfer Learning 10k labeled 데이터로 classifier 학습. 이미지넷처럼 ResNet-50을 사용하고 싶으나 모델이 너무 커 작은 데이터에 쉽게 오버피팅 ImageNet에서 pretrained된 ResNet-50을 load하고 10k data로 fine-tuning → 성능 향상 Language Model need to converting word to vectors one hot encoding Scaled poorly with vocab size Very high-dimensional sparse vectors → NN operations work poorly Violates what we know about word similarity (ex. dist..