chunked/nltk_chunking/model_doc_bit/chunk_3.txt

The abstract from the paper is the following:
Transfer of pre-trained representations improves sample efficiency and simplifies hyperparameter tuning when training deep neural networks for vision.