Overcoming Catastrophic Forgetting in Neural Networks. Kirkpatrick, J., Pascanu, R., Rabinowitz, N., Veness, J., Desjardins, G., Rusu, A. A., Milan, K., Quan, J., Ramalho, T., Grabska-Barwinska, A., Hassabis, D., Clopath, C., Kumaran, D., & Hadsell, R. 114(13):201611835–3526. ![Overcoming Catastrophic Forgetting in Neural Networks [link]](https://bibbase.org/img/filetypes/link.svg "link")
Paper doi abstract bibtex [Significance] Deep neural networks are currently the most successful machine-learning technique for solving a variety of tasks, including language translation, image classification, and image generation. One weakness of such models is that, unlike humans, they are unable to learn multiple tasks sequentially. In this work we propose a practical solution to train such models sequentially by protecting the weights important for previous tasks. This approach, inspired by synaptic consolidation in neuroscience, enables state of the art results on multiple reinforcement learning problems experienced sequentially. [Abstract] The ability to learn tasks in a sequential fashion is crucial to the development of artificial intelligence. Until now neural networks have not been capable of this and it has been widely thought that catastrophic forgetting is an inevitable feature of connectionist models. We show that it is possible to overcome this limitation and train networks that can maintain expertise on tasks that they have not experienced for a long time. Our approach remembers old tasks by selectively slowing down learning on the weights important for those tasks. We demonstrate our approach is scalable and effective by solving a set of classification tasks based on a hand-written digit dataset and by learning several Atari 2600 games sequentially.
@article{kirkpatrickOvercomingCatastrophicForgetting2017,
title = {Overcoming Catastrophic Forgetting in Neural Networks},
author = {Kirkpatrick, James and Pascanu, Razvan and Rabinowitz, Neil and Veness, Joel and Desjardins, Guillaume and Rusu, Andrei A. and Milan, Kieran and Quan, John and Ramalho, Tiago and Grabska-Barwinska, Agnieszka and Hassabis, Demis and Clopath, Claudia and Kumaran, Dharshan and Hadsell, Raia},
date = {2017-03},
journaltitle = {Proceedings of the National Academy of Sciences},
volume = {114},
pages = {201611835--3526},
issn = {1091-6490},
doi = {10.1073/pnas.1611835114},
url = {https://doi.org/10.1073/pnas.1611835114},
abstract = {[Significance]
Deep neural networks are currently the most successful machine-learning technique for solving a variety of tasks, including language translation, image classification, and image generation. One weakness of such models is that, unlike humans, they are unable to learn multiple tasks sequentially. In this work we propose a practical solution to train such models sequentially by protecting the weights important for previous tasks. This approach, inspired by synaptic consolidation in neuroscience, enables state of the art results on multiple reinforcement learning problems experienced sequentially. [Abstract]
The ability to learn tasks in a sequential fashion is crucial to the development of artificial intelligence. Until now neural networks have not been capable of this and it has been widely thought that catastrophic forgetting is an inevitable feature of connectionist models. We show that it is possible to overcome this limitation and train networks that can maintain expertise on tasks that they have not experienced for a long time. Our approach remembers old tasks by selectively slowing down learning on the weights important for those tasks. We demonstrate our approach is scalable and effective by solving a set of classification tasks based on a hand-written digit dataset and by learning several Atari 2600 games sequentially.},
keywords = {*imported-from-citeulike-INRMM,~INRMM-MiD:c-14311063,artificial-neural-networks,deep-machine-learning,memory,multiplicity,state-shift,system-catastrophe},
number = {13}
}
Downloads: 0
{"_id":"rDe2EquLkQAMTxNCF","bibbaseid":"kirkpatrick-pascanu-rabinowitz-veness-desjardins-rusu-milan-quan-etal-overcomingcatastrophicforgettinginneuralnetworks","authorIDs":[],"author_short":["Kirkpatrick, J.","Pascanu, R.","Rabinowitz, N.","Veness, J.","Desjardins, G.","Rusu, A. A.","Milan, K.","Quan, J.","Ramalho, T.","Grabska-Barwinska, A.","Hassabis, D.","Clopath, C.","Kumaran, D.","Hadsell, R."],"bibdata":{"bibtype":"article","type":"article","title":"Overcoming Catastrophic Forgetting in Neural Networks","author":[{"propositions":[],"lastnames":["Kirkpatrick"],"firstnames":["James"],"suffixes":[]},{"propositions":[],"lastnames":["Pascanu"],"firstnames":["Razvan"],"suffixes":[]},{"propositions":[],"lastnames":["Rabinowitz"],"firstnames":["Neil"],"suffixes":[]},{"propositions":[],"lastnames":["Veness"],"firstnames":["Joel"],"suffixes":[]},{"propositions":[],"lastnames":["Desjardins"],"firstnames":["Guillaume"],"suffixes":[]},{"propositions":[],"lastnames":["Rusu"],"firstnames":["Andrei","A."],"suffixes":[]},{"propositions":[],"lastnames":["Milan"],"firstnames":["Kieran"],"suffixes":[]},{"propositions":[],"lastnames":["Quan"],"firstnames":["John"],"suffixes":[]},{"propositions":[],"lastnames":["Ramalho"],"firstnames":["Tiago"],"suffixes":[]},{"propositions":[],"lastnames":["Grabska-Barwinska"],"firstnames":["Agnieszka"],"suffixes":[]},{"propositions":[],"lastnames":["Hassabis"],"firstnames":["Demis"],"suffixes":[]},{"propositions":[],"lastnames":["Clopath"],"firstnames":["Claudia"],"suffixes":[]},{"propositions":[],"lastnames":["Kumaran"],"firstnames":["Dharshan"],"suffixes":[]},{"propositions":[],"lastnames":["Hadsell"],"firstnames":["Raia"],"suffixes":[]}],"date":"2017-03","journaltitle":"Proceedings of the National Academy of Sciences","volume":"114","pages":"201611835–3526","issn":"1091-6490","doi":"10.1073/pnas.1611835114","url":"https://doi.org/10.1073/pnas.1611835114","abstract":"[Significance] Deep neural networks are currently the most successful machine-learning technique for solving a variety of tasks, including language translation, image classification, and image generation. One weakness of such models is that, unlike humans, they are unable to learn multiple tasks sequentially. In this work we propose a practical solution to train such models sequentially by protecting the weights important for previous tasks. This approach, inspired by synaptic consolidation in neuroscience, enables state of the art results on multiple reinforcement learning problems experienced sequentially. [Abstract] The ability to learn tasks in a sequential fashion is crucial to the development of artificial intelligence. Until now neural networks have not been capable of this and it has been widely thought that catastrophic forgetting is an inevitable feature of connectionist models. We show that it is possible to overcome this limitation and train networks that can maintain expertise on tasks that they have not experienced for a long time. Our approach remembers old tasks by selectively slowing down learning on the weights important for those tasks. We demonstrate our approach is scalable and effective by solving a set of classification tasks based on a hand-written digit dataset and by learning several Atari 2600 games sequentially.","keywords":"*imported-from-citeulike-INRMM,~INRMM-MiD:c-14311063,artificial-neural-networks,deep-machine-learning,memory,multiplicity,state-shift,system-catastrophe","number":"13","bibtex":"@article{kirkpatrickOvercomingCatastrophicForgetting2017,\n title = {Overcoming Catastrophic Forgetting in Neural Networks},\n author = {Kirkpatrick, James and Pascanu, Razvan and Rabinowitz, Neil and Veness, Joel and Desjardins, Guillaume and Rusu, Andrei A. and Milan, Kieran and Quan, John and Ramalho, Tiago and Grabska-Barwinska, Agnieszka and Hassabis, Demis and Clopath, Claudia and Kumaran, Dharshan and Hadsell, Raia},\n date = {2017-03},\n journaltitle = {Proceedings of the National Academy of Sciences},\n volume = {114},\n pages = {201611835--3526},\n issn = {1091-6490},\n doi = {10.1073/pnas.1611835114},\n url = {https://doi.org/10.1073/pnas.1611835114},\n abstract = {[Significance]\n\nDeep neural networks are currently the most successful machine-learning technique for solving a variety of tasks, including language translation, image classification, and image generation. One weakness of such models is that, unlike humans, they are unable to learn multiple tasks sequentially. In this work we propose a practical solution to train such models sequentially by protecting the weights important for previous tasks. This approach, inspired by synaptic consolidation in neuroscience, enables state of the art results on multiple reinforcement learning problems experienced sequentially. [Abstract]\n\nThe ability to learn tasks in a sequential fashion is crucial to the development of artificial intelligence. Until now neural networks have not been capable of this and it has been widely thought that catastrophic forgetting is an inevitable feature of connectionist models. We show that it is possible to overcome this limitation and train networks that can maintain expertise on tasks that they have not experienced for a long time. Our approach remembers old tasks by selectively slowing down learning on the weights important for those tasks. We demonstrate our approach is scalable and effective by solving a set of classification tasks based on a hand-written digit dataset and by learning several Atari 2600 games sequentially.},\n keywords = {*imported-from-citeulike-INRMM,~INRMM-MiD:c-14311063,artificial-neural-networks,deep-machine-learning,memory,multiplicity,state-shift,system-catastrophe},\n number = {13}\n}\n\n","author_short":["Kirkpatrick, J.","Pascanu, R.","Rabinowitz, N.","Veness, J.","Desjardins, G.","Rusu, A. A.","Milan, K.","Quan, J.","Ramalho, T.","Grabska-Barwinska, A.","Hassabis, D.","Clopath, C.","Kumaran, D.","Hadsell, R."],"key":"kirkpatrickOvercomingCatastrophicForgetting2017","id":"kirkpatrickOvercomingCatastrophicForgetting2017","bibbaseid":"kirkpatrick-pascanu-rabinowitz-veness-desjardins-rusu-milan-quan-etal-overcomingcatastrophicforgettinginneuralnetworks","role":"author","urls":{"Paper":"https://doi.org/10.1073/pnas.1611835114"},"keyword":["*imported-from-citeulike-INRMM","~INRMM-MiD:c-14311063","artificial-neural-networks","deep-machine-learning","memory","multiplicity","state-shift","system-catastrophe"],"downloads":0},"bibtype":"article","biburl":"https://tmpfiles.org/dl/58794/INRMM.bib","creationDate":"2020-07-02T22:41:11.170Z","downloads":0,"keywords":["*imported-from-citeulike-inrmm","~inrmm-mid:c-14311063","artificial-neural-networks","deep-machine-learning","memory","multiplicity","state-shift","system-catastrophe"],"search_terms":["overcoming","catastrophic","forgetting","neural","networks","kirkpatrick","pascanu","rabinowitz","veness","desjardins","rusu","milan","quan","ramalho","grabska-barwinska","hassabis","clopath","kumaran","hadsell"],"title":"Overcoming Catastrophic Forgetting in Neural Networks","year":null,"dataSources":["DXuKbcZTirdigFKPF"]}