Louis Fournier, Stephane Rivaud, Eugene Belilovsky, Michael Eickenberg, Edouard Oyallon
{"title":"前向梯度能匹配反向传播吗?","authors":"Louis Fournier, Stephane Rivaud, Eugene Belilovsky, Michael Eickenberg, Edouard Oyallon","doi":"10.48550/arXiv.2306.06968","DOIUrl":null,"url":null,"abstract":"Forward Gradients - the idea of using directional derivatives in forward differentiation mode - have recently been shown to be utilizable for neural network training while avoiding problems generally associated with backpropagation gradient computation, such as locking and memorization requirements. The cost is the requirement to guess the step direction, which is hard in high dimensions. While current solutions rely on weighted averages over isotropic guess vector distributions, we propose to strongly bias our gradient guesses in directions that are much more promising, such as feedback obtained from small, local auxiliary networks. For a standard computer vision neural network, we conduct a rigorous study systematically covering a variety of combinations of gradient targets and gradient guesses, including those previously presented in the literature. We find that using gradients obtained from a local loss as a candidate direction drastically improves on random noise in Forward Gradient methods.","PeriodicalId":74529,"journal":{"name":"Proceedings of the ... International Conference on Machine Learning. International Conference on Machine Learning","volume":"24 1","pages":"10249-10264"},"PeriodicalIF":0.0000,"publicationDate":"2023-06-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Can Forward Gradient Match Backpropagation?\",\"authors\":\"Louis Fournier, Stephane Rivaud, Eugene Belilovsky, Michael Eickenberg, Edouard Oyallon\",\"doi\":\"10.48550/arXiv.2306.06968\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Forward Gradients - the idea of using directional derivatives in forward differentiation mode - have recently been shown to be utilizable for neural network training while avoiding problems generally associated with backpropagation gradient computation, such as locking and memorization requirements. The cost is the requirement to guess the step direction, which is hard in high dimensions. While current solutions rely on weighted averages over isotropic guess vector distributions, we propose to strongly bias our gradient guesses in directions that are much more promising, such as feedback obtained from small, local auxiliary networks. For a standard computer vision neural network, we conduct a rigorous study systematically covering a variety of combinations of gradient targets and gradient guesses, including those previously presented in the literature. We find that using gradients obtained from a local loss as a candidate direction drastically improves on random noise in Forward Gradient methods.\",\"PeriodicalId\":74529,\"journal\":{\"name\":\"Proceedings of the ... International Conference on Machine Learning. International Conference on Machine Learning\",\"volume\":\"24 1\",\"pages\":\"10249-10264\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2023-06-12\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Proceedings of the ... International Conference on Machine Learning. International Conference on Machine Learning\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.48550/arXiv.2306.06968\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Proceedings of the ... International Conference on Machine Learning. International Conference on Machine Learning","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.48550/arXiv.2306.06968","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Forward Gradients - the idea of using directional derivatives in forward differentiation mode - have recently been shown to be utilizable for neural network training while avoiding problems generally associated with backpropagation gradient computation, such as locking and memorization requirements. The cost is the requirement to guess the step direction, which is hard in high dimensions. While current solutions rely on weighted averages over isotropic guess vector distributions, we propose to strongly bias our gradient guesses in directions that are much more promising, such as feedback obtained from small, local auxiliary networks. For a standard computer vision neural network, we conduct a rigorous study systematically covering a variety of combinations of gradient targets and gradient guesses, including those previously presented in the literature. We find that using gradients obtained from a local loss as a candidate direction drastically improves on random noise in Forward Gradient methods.