@article{ramachandran_unsupervised_2017,
	title = {Unsupervised {Pretraining} for {Sequence} to {Sequence} {Learning}},
	url = {http://aclweb.org/anthology/D17-1039},
	doi = {10.18653/v1/D17-1039},
	abstract = {This work presents a general unsupervised learning method to improve the accuracy of sequence to sequence (seq2seq) models. In our method, the weights of the encoder and decoder of a seq2seq model are initialized with the pretrained weights of two language models and then fine-tuned with labeled data. We apply this method to challenging benchmarks in machine translation and abstractive summarization and find that it significantly improves the subsequent supervised models. Our main result is that pretraining improves the generalization of seq2seq models. We achieve state-of-the-art results on the WMT English→German task, surpassing a range of methods using both phrase-based machine translation and neural machine translation. Our method achieves a significant improvement of 1.3 BLEU from th previous best models on both WMT’14 and WMT’15 English→German. We also conduct human evaluations on abstractive summarization and find that our method outperforms a purely supervised learning baseline in a statistically significant manner.},
	language = {en},
	urldate = {2024-04-13},
	journal = {Proceedings of the 2017 Conference on Empirical Methods in Natural           Language Processing},
	author = {Ramachandran, Prajit and Liu, Peter and Le, Quoc},
	year = {2017},
	note = {Conference Name: Proceedings of the 2017 Conference on Empirical Methods in Natural
          Language Processing
Place: Copenhagen, Denmark
Publisher: Association for Computational Linguistics},
	pages = {383--391},
}

{"_id":"KpCRsrGeK4xoikXFs","bibbaseid":"ramachandran-liu-le-unsupervisedpretrainingforsequencetosequencelearning-2017","author_short":["Ramachandran, P.","Liu, P.","Le, Q."],"bibdata":{"bibtype":"article","type":"article","title":"Unsupervised Pretraining for Sequence to Sequence Learning","url":"http://aclweb.org/anthology/D17-1039","doi":"10.18653/v1/D17-1039","abstract":"This work presents a general unsupervised learning method to improve the accuracy of sequence to sequence (seq2seq) models. In our method, the weights of the encoder and decoder of a seq2seq model are initialized with the pretrained weights of two language models and then fine-tuned with labeled data. We apply this method to challenging benchmarks in machine translation and abstractive summarization and find that it significantly improves the subsequent supervised models. Our main result is that pretraining improves the generalization of seq2seq models. We achieve state-of-the-art results on the WMT English→German task, surpassing a range of methods using both phrase-based machine translation and neural machine translation. Our method achieves a significant improvement of 1.3 BLEU from th previous best models on both WMT’14 and WMT’15 English→German. We also conduct human evaluations on abstractive summarization and find that our method outperforms a purely supervised learning baseline in a statistically significant manner.","language":"en","urldate":"2024-04-13","journal":"Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing","author":[{"propositions":[],"lastnames":["Ramachandran"],"firstnames":["Prajit"],"suffixes":[]},{"propositions":[],"lastnames":["Liu"],"firstnames":["Peter"],"suffixes":[]},{"propositions":[],"lastnames":["Le"],"firstnames":["Quoc"],"suffixes":[]}],"year":"2017","note":"Conference Name: Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing Place: Copenhagen, Denmark Publisher: Association for Computational Linguistics","pages":"383–391","bibtex":"@article{ramachandran_unsupervised_2017,\n\ttitle = {Unsupervised {Pretraining} for {Sequence} to {Sequence} {Learning}},\n\turl = {http://aclweb.org/anthology/D17-1039},\n\tdoi = {10.18653/v1/D17-1039},\n\tabstract = {This work presents a general unsupervised learning method to improve the accuracy of sequence to sequence (seq2seq) models. In our method, the weights of the encoder and decoder of a seq2seq model are initialized with the pretrained weights of two language models and then fine-tuned with labeled data. We apply this method to challenging benchmarks in machine translation and abstractive summarization and find that it significantly improves the subsequent supervised models. Our main result is that pretraining improves the generalization of seq2seq models. We achieve state-of-the-art results on the WMT English→German task, surpassing a range of methods using both phrase-based machine translation and neural machine translation. Our method achieves a significant improvement of 1.3 BLEU from th previous best models on both WMT’14 and WMT’15 English→German. We also conduct human evaluations on abstractive summarization and find that our method outperforms a purely supervised learning baseline in a statistically significant manner.},\n\tlanguage = {en},\n\turldate = {2024-04-13},\n\tjournal = {Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing},\n\tauthor = {Ramachandran, Prajit and Liu, Peter and Le, Quoc},\n\tyear = {2017},\n\tnote = {Conference Name: Proceedings of the 2017 Conference on Empirical Methods in Natural\n Language Processing\nPlace: Copenhagen, Denmark\nPublisher: Association for Computational Linguistics},\n\tpages = {383--391},\n}\n\n\n\n\n\n\n\n","author_short":["Ramachandran, P.","Liu, P.","Le, Q."],"key":"ramachandran_unsupervised_2017","id":"ramachandran_unsupervised_2017","bibbaseid":"ramachandran-liu-le-unsupervisedpretrainingforsequencetosequencelearning-2017","role":"author","urls":{"Paper":"http://aclweb.org/anthology/D17-1039"},"metadata":{"authorlinks":{}}},"bibtype":"article","biburl":"https://bibbase.org/zotero/abhishek-p","dataSources":["h7kKWXpJh2iaX92T5"],"keywords":[],"search_terms":["unsupervised","pretraining","sequence","sequence","learning","ramachandran","liu","le"],"title":"Unsupervised Pretraining for Sequence to Sequence Learning","year":2017}