@misc{ruder2017learning,
  abstract = {Multi-task learning is motivated by the observation that humans bring to bear
what they know about related problems when solving new ones. Similarly, deep
neural networks can profit from related tasks by sharing parameters with other
networks. However, humans do not consciously decide to transfer knowledge
between tasks. In Natural Language Processing (NLP), it is hard to predict if
sharing will lead to improvements, particularly if tasks are only loosely
related. To overcome this, we introduce Sluice Networks, a general framework
for multi-task learning where trainable parameters control the amount of
sharing. Our framework generalizes previous proposals in enabling sharing of
all combinations of subspaces, layers, and skip connections. We perform
experiments on three task pairs, and across seven different domains, using data
from OntoNotes 5.0, and achieve up to 15% average error reductions over common
approaches to multi-task learning. We show that a) label entropy is predictive
of gains in sluice networks, confirming findings for hard parameter sharing and
b) while sluice networks easily fit noise, they are robust across domains in
practice.},
  added-at = {2018-04-27T11:50:18.000+0200},
  author = {Ruder, Sebastian and Bingel, Joachim and Augenstein, Isabelle and Søgaard, Anders},
  biburl = {https://www.bibsonomy.org/bibtex/2ee6714f7d4015de7a48f803eb8123adf/dallmann},
  description = {Learning what to share between loosely related tasks},
  interhash = {a1276ad70b9022e5a753453ee0b71005},
  intrahash = {ee6714f7d4015de7a48f803eb8123adf},
  keywords = {multitask nlp deep_learning},
  note = {cite arxiv:1705.08142Comment: 12 pages, 3 figures, 6 tables},
  timestamp = {2018-04-27T11:50:18.000+0200},
  title = {Learning what to share between loosely related tasks},
  url = {http://arxiv.org/abs/1705.08142},
  year = 2017
}

{"_id":"jiG3QPzZGirEctxTA","bibbaseid":"ruder-bingel-augenstein-sgaard-learningwhattosharebetweenlooselyrelatedtasks-2017","author_short":["Ruder, S.","Bingel, J.","Augenstein, I.","Søgaard, A."],"bibdata":{"bibtype":"misc","type":"misc","abstract":"Multi-task learning is motivated by the observation that humans bring to bear what they know about related problems when solving new ones. Similarly, deep neural networks can profit from related tasks by sharing parameters with other networks. However, humans do not consciously decide to transfer knowledge between tasks. In Natural Language Processing (NLP), it is hard to predict if sharing will lead to improvements, particularly if tasks are only loosely related. To overcome this, we introduce Sluice Networks, a general framework for multi-task learning where trainable parameters control the amount of sharing. Our framework generalizes previous proposals in enabling sharing of all combinations of subspaces, layers, and skip connections. We perform experiments on three task pairs, and across seven different domains, using data from OntoNotes 5.0, and achieve up to 15% average error reductions over common approaches to multi-task learning. We show that a) label entropy is predictive of gains in sluice networks, confirming findings for hard parameter sharing and b) while sluice networks easily fit noise, they are robust across domains in practice.","added-at":"2018-04-27T11:50:18.000+0200","author":[{"propositions":[],"lastnames":["Ruder"],"firstnames":["Sebastian"],"suffixes":[]},{"propositions":[],"lastnames":["Bingel"],"firstnames":["Joachim"],"suffixes":[]},{"propositions":[],"lastnames":["Augenstein"],"firstnames":["Isabelle"],"suffixes":[]},{"propositions":[],"lastnames":["Søgaard"],"firstnames":["Anders"],"suffixes":[]}],"biburl":"https://www.bibsonomy.org/bibtex/2ee6714f7d4015de7a48f803eb8123adf/dallmann","description":"Learning what to share between loosely related tasks","interhash":"a1276ad70b9022e5a753453ee0b71005","intrahash":"ee6714f7d4015de7a48f803eb8123adf","keywords":"multitask nlp deep_learning","note":"cite arxiv:1705.08142Comment: 12 pages, 3 figures, 6 tables","timestamp":"2018-04-27T11:50:18.000+0200","title":"Learning what to share between loosely related tasks","url":"http://arxiv.org/abs/1705.08142","year":"2017","bibtex":"@misc{ruder2017learning,\n abstract = {Multi-task learning is motivated by the observation that humans bring to bear\r\nwhat they know about related problems when solving new ones. Similarly, deep\r\nneural networks can profit from related tasks by sharing parameters with other\r\nnetworks. However, humans do not consciously decide to transfer knowledge\r\nbetween tasks. In Natural Language Processing (NLP), it is hard to predict if\r\nsharing will lead to improvements, particularly if tasks are only loosely\r\nrelated. To overcome this, we introduce Sluice Networks, a general framework\r\nfor multi-task learning where trainable parameters control the amount of\r\nsharing. Our framework generalizes previous proposals in enabling sharing of\r\nall combinations of subspaces, layers, and skip connections. We perform\r\nexperiments on three task pairs, and across seven different domains, using data\r\nfrom OntoNotes 5.0, and achieve up to 15% average error reductions over common\r\napproaches to multi-task learning. We show that a) label entropy is predictive\r\nof gains in sluice networks, confirming findings for hard parameter sharing and\r\nb) while sluice networks easily fit noise, they are robust across domains in\r\npractice.},\n added-at = {2018-04-27T11:50:18.000+0200},\n author = {Ruder, Sebastian and Bingel, Joachim and Augenstein, Isabelle and Søgaard, Anders},\n biburl = {https://www.bibsonomy.org/bibtex/2ee6714f7d4015de7a48f803eb8123adf/dallmann},\n description = {Learning what to share between loosely related tasks},\n interhash = {a1276ad70b9022e5a753453ee0b71005},\n intrahash = {ee6714f7d4015de7a48f803eb8123adf},\n keywords = {multitask nlp deep_learning},\n note = {cite arxiv:1705.08142Comment: 12 pages, 3 figures, 6 tables},\n timestamp = {2018-04-27T11:50:18.000+0200},\n title = {Learning what to share between loosely related tasks},\n url = {http://arxiv.org/abs/1705.08142},\n year = 2017\n}\n\n","author_short":["Ruder, S.","Bingel, J.","Augenstein, I.","Søgaard, A."],"key":"ruder2017learning","id":"ruder2017learning","bibbaseid":"ruder-bingel-augenstein-sgaard-learningwhattosharebetweenlooselyrelatedtasks-2017","role":"author","urls":{"Paper":"http://arxiv.org/abs/1705.08142"},"keyword":["multitask nlp deep_learning"],"metadata":{"authorlinks":{}},"html":""},"bibtype":"misc","biburl":"http://www.bibsonomy.org/bib/author/Ruder?items=1000","dataSources":["JFLn7yTRgDTDaR4Am"],"keywords":["multitask nlp deep_learning"],"search_terms":["learning","share","between","loosely","related","tasks","ruder","bingel","augenstein","søgaard"],"title":"Learning what to share between loosely related tasks","year":2017}