After AutoAugment, RandAugment, AutoML/NasNet/EfficientNet/EfficientDet…. Quoc V. Le (and his colleagues at Google Brain) strike again to automate another part of deep learning models and training :
\nAbstract
\n\n\n\nNeural networks are often over-parameterized and hence benefit from aggressive regularization. Conventional regularization methods, such as Dropout or weight decay, do not leverage the structures of the network's inputs and hidden states. As a result, these conventional methods are less effective than methods that leverage the structures, such as SpatialDropout and DropBlock, which randomly drop the values at certain contiguous areas in the hidden states and setting them to zero. Although the locations of dropout areas random, the patterns of SpatialDropout and DropBlock are manually designed and fixed. Here we propose to learn the dropout patterns. In our method, a controller learns to generate a dropout pattern at every channel and layer of a target network, such as a ConvNet or a Transformer. The target network is then trained with the dropout pattern, and its resulting validation performance is used as a signal for the controller to learn from. We show that this method works well for both image recognition on CIFAR-10 and ImageNet, as well as language modeling on Penn Treebank and WikiText-2. The learned dropout patterns also transfers to different tasks and datasets, such as from language model on Penn Treebank to Engligh-French translation on WMT 2014
\n
Code link (code will be available soon)
", "messages": [ { "id": "1142461", "postDate": "01/07/2021 11:57:38", "content": "After AutoAugment, RandAugment, AutoML/NasNet/EfficientNet/EfficientDet…. Quoc V. Le (and his colleagues at Google Brain) strike again to automate another part of deep learning models and training :
\nAbstract
\n\n\n\nNeural networks are often over-parameterized and hence benefit from aggressive regularization. Conventional regularization methods, such as Dropout or weight decay, do not leverage the structures of the network's inputs and hidden states. As a result, these conventional methods are less effective than methods that leverage the structures, such as SpatialDropout and DropBlock, which randomly drop the values at certain contiguous areas in the hidden states and setting them to zero. Although the locations of dropout areas random, the patterns of SpatialDropout and DropBlock are manually designed and fixed. Here we propose to learn the dropout patterns. In our method, a controller learns to generate a dropout pattern at every channel and layer of a target network, such as a ConvNet or a Transformer. The target network is then trained with the dropout pattern, and its resulting validation performance is used as a signal for the controller to learn from. We show that this method works well for both image recognition on CIFAR-10 and ImageNet, as well as language modeling on Penn Treebank and WikiText-2. The learned dropout patterns also transfers to different tasks and datasets, such as from language model on Penn Treebank to Engligh-French translation on WMT 2014
\n
Code link (code will be available soon)
", "rawMarkdown": "After AutoAugment, RandAugment, AutoML/NasNet/EfficientNet/EfficientDet.... Quoc V. Le (and his colleagues at Google Brain) strike again to automate another part of deep learning models and training : \n\n**Abstract** \n\n> Neural networks are often over-parameterized and hence benefit from aggressive regularization. Conventional regularization methods, such as Dropout or weight decay, do not leverage the structures of the network's inputs and hidden states. As a result, these conventional methods are less effective than methods that leverage the structures, such as SpatialDropout and DropBlock, which randomly drop the values at certain contiguous areas in the hidden states and setting them to zero. Although the locations of dropout areas random, the patterns of SpatialDropout and DropBlock are manually designed and fixed. Here we propose to learn the dropout patterns. In our method, a controller learns to generate a dropout pattern at every channel and layer of a target network, such as a ConvNet or a Transformer. The target network is then trained with the dropout pattern, and its resulting validation performance is used as a signal for the controller to learn from. We show that this method works well for both image recognition on CIFAR-10 and ImageNet, as well as language modeling on Penn Treebank and WikiText-2. The learned dropout patterns also transfers to different tasks and datasets, such as from language model on Penn Treebank to Engligh-French translation on WMT 2014\n\n[Paper link](https://arxiv.org/pdf/2101.01761.pdf)\n\n[Code link](https://github.com/google-research/google-research/tree/master/auto_dropout) (code will be available soon)", "votes": null }, { "id": "1143392", "postDate": "01/07/2021 22:04:24", "content": "Looks like everything is getting automated!
", "rawMarkdown": "Looks like everything is getting automated!", "votes": null }, { "id": "1144218", "postDate": "01/08/2021 10:20:25", "content": "@serigne Know any good paper to read about AutoML?
", "rawMarkdown": "serigne Know any good paper to read about AutoML?", "votes": null } ], "comments": [ { "id": 1143392, "author_name": "debarshichanda", "author_url": "", "post_date": "01/07/2021 22:04:24", "content": "Looks like everything is getting automated!
", "votes": null, "replies": [] }, { "id": 1144218, "author_name": "prvnkmr", "author_url": "", "post_date": "01/08/2021 10:20:25", "content": "@serigne Know any good paper to read about AutoML?
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1142461": "After AutoAugment, RandAugment, AutoML/NasNet/EfficientNet/EfficientDet.... Quoc V. Le (and his colleagues at Google Brain) strike again to automate another part of deep learning models and training : \n\n**Abstract** \n\n> Neural networks are often over-parameterized and hence benefit from aggressive regularization. Conventional regularization methods, such as Dropout or weight decay, do not leverage the structures of the network's inputs and hidden states. As a result, these conventional methods are less effective than methods that leverage the structures, such as SpatialDropout and DropBlock, which randomly drop the values at certain contiguous areas in the hidden states and setting them to zero. Although the locations of dropout areas random, the patterns of SpatialDropout and DropBlock are manually designed and fixed. Here we propose to learn the dropout patterns. In our method, a controller learns to generate a dropout pattern at every channel and layer of a target network, such as a ConvNet or a Transformer. The target network is then trained with the dropout pattern, and its resulting validation performance is used as a signal for the controller to learn from. We show that this method works well for both image recognition on CIFAR-10 and ImageNet, as well as language modeling on Penn Treebank and WikiText-2. The learned dropout patterns also transfers to different tasks and datasets, such as from language model on Penn Treebank to Engligh-French translation on WMT 2014\n\n[Paper link](https://arxiv.org/pdf/2101.01761.pdf)\n\n[Code link](https://github.com/google-research/google-research/tree/master/auto_dropout) (code will be available soon)", "1143392": "Looks like everything is getting automated!", "1144218": "serigne Know any good paper to read about AutoML?" }, "source": "meta" }