{"id":680,"date":"2015-07-11T01:36:08","date_gmt":"2015-07-11T05:36:08","guid":{"rendered":"http:\/\/zhanxw.com\/blog\/?p=680"},"modified":"2015-07-11T01:51:26","modified_gmt":"2015-07-11T05:51:26","slug":"%e6%b7%b1%e5%ba%a6%e5%ad%a6%e4%b9%a0mnist%e6%95%b0%e6%8d%ae","status":"publish","type":"post","link":"https:\/\/zhanxw.com\/blog\/2015\/07\/%e6%b7%b1%e5%ba%a6%e5%ad%a6%e4%b9%a0mnist%e6%95%b0%e6%8d%ae\/","title":{"rendered":"\u6df1\u5ea6\u5b66\u4e60MNIST\u6570\u636e"},"content":{"rendered":"<p>\u6df1\u5ea6\u5b66\u4e60MNIST\u6570\u636e<\/p>\n<p>Deep learning on MNIST data<\/p>\n<p>\u6700\u8fd1\u8bd5\u4e86\u4e00\u4e0bDeep Learning\u7b97\u6cd5\uff0c\u5728Kaggle\u7684<a href=\"https:\/\/www.kaggle.com\/c\/digit-recognizer\">Digit Recognizer<\/a>\u6392\u884c\u699c\u4e0a\u76ee\u524d\u639219\u540d\u3002<\/p>\n<p>\u7b80\u5355\u8bf4\u8bf4\u7ecf\u9a8c\u3002<\/p>\n<p>\u4ece\u5de5\u5177\u4e0a\u8bb2\uff0c\u5efa\u8bae\u5148\u7528<a href=\"http:\/\/keras.io\/\">Keras<\/a>\u6765\u505a\u539f\u578b\uff08\u4ee3\u7801\u7b80\u5355\uff0c\u5f88\u5bb9\u6613\u6784\u5efa\u590d\u6742\u7684\u7f51\u7edc\uff0c\u7f3a\u70b9\u662f\u592a\u5360\u5185\u5b58\uff0f\u663e\u5b58\uff09\uff0c\u518d\u7528<a href=\"http:\/\/caffe.berkeleyvision.org\/\">Caffe<\/a>\u505a\u66f4\u7ec6\u81f4\u7684\u8c03\u8282\uff08\u8ba1\u7b97\u5feb\uff0c\u5360\u5185\u5b58\uff0f\u663e\u5b58\u5c11\uff0c\u4f46\u7528\u8d77\u6765\u590d\u6742\uff0c\u6bd4\u5982\u4e0d\u652f\u6301RMSProp\uff0c\u9700\u8981\u624b\u52a8\u628a\u6570\u636e\u5206\u6210Training\/Testing\uff0c\u8981\u624b\u52a8\u7528proto buffer\u6765\u6784\u5efa\u795e\u7ecf\u7f51\u7edc\uff09\u3002\u6b64\u5916\uff0c\u7528GPU\u52a0\u901f\u662f\u5fc5\u987b\u7684\uff0c\u6211\u89c2\u5bdf\u5230GPU\u901f\u5ea6\u81f3\u5c11\u662fCPU\u901f\u5ea6\u768460\u500d\u3002<\/p>\n<p>\u4ece\u6a21\u578b\u4e0a\u8bb2\uff0c\u57fa\u672c\u4e0a\u5efa\u7acb\u4e0d\u540c\u7684Convolution Network\u3002\u5957\u8def\u4f3c\u4e4e\u90fd\u662f\u91cd\u590d\u8fd9\u4e2a\u7ed3\u6784\uff1aConvolution Layer + ReLU Layer + MaxPooling Layer\u3002\u6a21\u578b\u53ef\u4ee5\u66f4Deep \uff08\u5c42\u6570\u66f4\u591a\uff09\uff0c\u6bd4\u5982VGG\u7f51\u7edc\uff0c\u4f46\u662f\u4ee3\u4ef7\u662f<a href=\"http:\/\/cs231n.github.io\/convolutional-networks\/#case\">\u53c2\u6570\u6570\u91cf\u591a\uff0c\u8ba1\u7b97\u91cf\u592a\u5927<\/a>\u3002\u5bf9\u4e8eMNIST\u8fd9\u4e2a\u6574\u7406\u597d\uff08Pre-processed\uff09\u7684\u6570\u636e\uff0c\u8fd9\u4e9b\u7f51\u7edc\u7684\u6548\u679c\u5f88\u76f8\u4f3c\uff0c\u4e00\u822c\u9ed8\u8ba4\u8bbe\u7f6e\u90fd\u80fd\u8fbe\u523098%\u6b63\u786e\u7387\u4ee5\u4e0a\uff08\u752842k\u6570\u636e\u8bad\u7ec3\uff0c28k\u6570\u636e\u6d4b\u8bd5\uff09\u3002\u5982\u679c\u8981\u66f4\u597d\u7684\u6210\u7ee9\uff0c\u57fa\u672c\u4e0a\u9700\u8981\u66f4\u591a\u7684\u8f93\u5165\u6570\u636e\uff08\u6bd4\u5982\u628a\u539f\u59cb\u7684\u56fe\u50cf\u4e0a\u4e0b\u5de6\u53f3\u79fb\u52a8\uff09\uff0c\u66f4\u6df1\u7684\u6a21\u578b\uff08\u6bd4\u5982\u589e\u52a0\u5c42\u6570\uff09\uff0c\u66f4\u591a\u7684\u6a21\u578b\uff08\u6bd4\u5982\u91cd\u590d5\u904d\uff0c\u518dEnsemble\u7ed3\u679c\uff09\uff0c\u8fd9\u6837\u4e00\u822c\u80fd\u7a33\u5b9a\u8fbe\u523099%\u4ee5\u4e0a\u7684\u6b63\u786e\u7387\u3002<\/p>\n<p>\u76ee\u524d\u80fd\u8fbe\u523099.486%\u6b63\u786e\u6027\u7684\u6a21\u578b\u662f\uff1a\u628a\u5df2\u6709\u768442k\u6570\u636e\u6309\u716780%, 20%\u7684\u6bd4\u4f8b\u5206\u6210training,testing\u6570\u636e\uff0c\u7136\u540e\u7528Lenet\uff08120 convolution layers [kernel = 5] + ReLU + MaxPooling + 200 convolution layer [kernel = 3] + ReLU + MaxPooling + InnerProduct [param = 200] + InnerProduct [param=10]), \u8fed\u4ee360,000\u6b21\u3002\u7136\u540e\u8fd9\u4e2a\u8fc7\u7a0b\u91cd\u590d15\u6b21\uff0c\u7136\u540e\u9009\u591a\u6570Ensemble\u3002<\/p>\n<p>\u6700\u540e\u653e\u51e0\u4e2a\u795e\u7ecf\u7f51\u7edc\u8bc6\u522b\u8d77\u6765\u8d39\u52b2\u7684\u56fe\u6848\u4f9b\u6b23\u8d4f\uff0c\u8fd8\u6709\u4e00\u4e2a\u4ecb\u7ecd\u6027\u8d28\u7684Slides\u3002<\/p>\n<p>&nbsp;<\/p>\n<figure id=\"attachment_681\" class=\"thumbnail wp-caption alignnone\" style=\"width: 967px\"><a href=\"http:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/Screen-Shot-2015-07-11-at-12.04.26-AM.png\"><img loading=\"lazy\" decoding=\"async\" class=\"size-full wp-image-681\" src=\"http:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/Screen-Shot-2015-07-11-at-12.04.26-AM.png\" alt=\"DigitRecognizerRank\" width=\"957\" height=\"869\" srcset=\"https:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/Screen-Shot-2015-07-11-at-12.04.26-AM.png 957w, https:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/Screen-Shot-2015-07-11-at-12.04.26-AM-300x272.png 300w, https:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/Screen-Shot-2015-07-11-at-12.04.26-AM-900x817.png 900w\" sizes=\"auto, (max-width: 957px) 100vw, 957px\" \/><\/a><figcaption class=\"caption wp-caption-text\">DigitRecognizerRank<\/figcaption><\/figure>\n<p><a href=\"http:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/probably-4.png\"><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-682\" src=\"http:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/probably-4.png\" alt=\"probably 4\" width=\"365\" height=\"61\" srcset=\"https:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/probably-4.png 365w, https:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/probably-4-300x50.png 300w\" sizes=\"auto, (max-width: 365px) 100vw, 365px\" \/><\/a> <a href=\"http:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/looks-like-9.png\"><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-683\" src=\"http:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/looks-like-9.png\" alt=\"looks like 9\" width=\"365\" height=\"66\" srcset=\"https:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/looks-like-9.png 365w, https:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/looks-like-9-300x54.png 300w\" sizes=\"auto, (max-width: 365px) 100vw, 365px\" \/><\/a> <a href=\"http:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/3-or-2.png\"><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-684\" src=\"http:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/3-or-2.png\" alt=\"3 or 2\" width=\"365\" height=\"64\" srcset=\"https:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/3-or-2.png 365w, https:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/3-or-2-300x53.png 300w\" sizes=\"auto, (max-width: 365px) 100vw, 365px\" \/><\/a> <a href=\"http:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/7-or-2.png\"><img loading=\"lazy\" decoding=\"async\" class=\"alignnone size-full wp-image-685\" src=\"http:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/7-or-2.png\" alt=\"7 or 2\" width=\"366\" height=\"65\" srcset=\"https:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/7-or-2.png 366w, https:\/\/zhanxw.com\/blog\/wp-content\/uploads\/2015\/07\/7-or-2-300x53.png 300w\" sizes=\"auto, (max-width: 366px) 100vw, 366px\" \/><\/a><\/p>\n<p><iframe loading=\"lazy\" src=\"\/\/www.slideshare.net\/slideshow\/embed_code\/key\/7znUkWBeTG1QPe\" width=\"425\" height=\"355\" frameborder=\"0\" marginwidth=\"0\" marginheight=\"0\" scrolling=\"no\" style=\"border:1px solid #CCC; border-width:1px; margin-bottom:5px; max-width: 100%;\" allowfullscreen> <\/iframe> <\/p>\n<div style=\"margin-bottom:5px\"> <strong> <a href=\"\/\/www.slideshare.net\/roelofp\/python-for-image-understanding-deep-learning-with-convolutional-neural-nets\" title=\"Python for Image Understanding: Deep Learning with Convolutional Neural Nets\" target=\"_blank\">Python for Image Understanding: Deep Learning with Convolutional Neural Nets<\/a> <\/strong> from <strong><a href=\"\/\/www.slideshare.net\/roelofp\" target=\"_blank\">Roelof Pieters<\/a><\/strong> <\/div>\n","protected":false},"excerpt":{"rendered":"<p>\u6df1\u5ea6\u5b66\u4e60MNIST\u6570\u636e Deep learning on MNIST data \u6700\u8fd1\u8bd5\u4e86\u4e00\u4e0bDeep Learning\u7b97\u6cd5\uff0c\u5728Kaggle\u7684Digit Recognizer\u6392\u884c\u699c\u4e0a\u76ee\u524d\u639219\u540d\u3002 \u7b80\u5355\u8bf4\u8bf4\u7ecf\u9a8c\u3002 \u4ece\u5de5\u5177\u4e0a\u8bb2\uff0c\u5efa\u8bae\u5148\u7528Keras\u6765\u505a\u539f\u578b\uff08\u4ee3\u7801\u7b80\u5355\uff0c\u5f88\u5bb9\u6613\u6784\u5efa\u590d\u6742\u7684\u7f51\u7edc\uff0c\u7f3a\u70b9\u662f\u592a\u5360\u5185\u5b58\uff0f\u663e\u5b58\uff09\uff0c\u518d\u7528Caffe\u505a\u66f4\u7ec6\u81f4\u7684\u8c03\u8282\uff08\u8ba1\u7b97\u5feb\uff0c\u5360\u5185\u5b58\uff0f\u663e\u5b58\u5c11\uff0c\u4f46\u7528\u8d77\u6765\u590d\u6742\uff0c\u6bd4\u5982\u4e0d\u652f\u6301RMSProp\uff0c\u9700\u8981\u624b\u52a8\u628a\u6570\u636e\u5206\u6210Training\/Testing\uff0c\u8981\u624b\u52a8\u7528proto buffer\u6765\u6784\u5efa\u795e\u7ecf\u7f51\u7edc\uff09\u3002\u6b64\u5916\uff0c\u7528GPU\u52a0\u901f\u662f\u5fc5\u987b\u7684\uff0c\u6211\u89c2\u5bdf\u5230GPU\u901f\u5ea6\u81f3\u5c11\u662fCPU\u901f\u5ea6\u768460\u500d\u3002 \u4ece\u6a21\u578b\u4e0a\u8bb2\uff0c\u57fa\u672c\u4e0a\u5efa\u7acb\u4e0d\u540c\u7684Convolution Network\u3002\u5957\u8def\u4f3c\u4e4e\u90fd\u662f\u91cd\u590d\u8fd9\u4e2a\u7ed3\u6784\uff1aConvolution Layer + ReLU Layer + MaxPooling Layer\u3002\u6a21\u578b\u53ef\u4ee5\u66f4Deep \uff08\u5c42\u6570\u66f4\u591a\uff09\uff0c\u6bd4\u5982VGG\u7f51\u7edc\uff0c\u4f46\u662f\u4ee3\u4ef7\u662f\u53c2\u6570\u6570\u91cf\u591a\uff0c\u8ba1\u7b97\u91cf\u592a\u5927\u3002\u5bf9\u4e8eMNIST\u8fd9\u4e2a\u6574\u7406\u597d\uff08Pre-processed\uff09\u7684\u6570\u636e\uff0c\u8fd9\u4e9b\u7f51\u7edc\u7684\u6548\u679c\u5f88\u76f8\u4f3c\uff0c\u4e00\u822c\u9ed8\u8ba4\u8bbe\u7f6e\u90fd\u80fd\u8fbe\u523098%\u6b63\u786e\u7387\u4ee5\u4e0a\uff08\u752842k\u6570\u636e\u8bad\u7ec3\uff0c28k\u6570\u636e\u6d4b\u8bd5\uff09\u3002\u5982\u679c\u8981\u66f4\u597d\u7684\u6210\u7ee9\uff0c\u57fa\u672c\u4e0a\u9700\u8981\u66f4\u591a\u7684\u8f93\u5165\u6570\u636e\uff08\u6bd4\u5982\u628a\u539f\u59cb\u7684\u56fe\u50cf\u4e0a\u4e0b\u5de6\u53f3\u79fb\u52a8\uff09\uff0c\u66f4\u6df1\u7684\u6a21\u578b\uff08\u6bd4\u5982\u589e\u52a0\u5c42\u6570\uff09\uff0c\u66f4\u591a\u7684\u6a21\u578b\uff08\u6bd4\u5982\u91cd\u590d5\u904d\uff0c\u518dEnsemble\u7ed3\u679c\uff09\uff0c\u8fd9\u6837\u4e00\u822c\u80fd\u7a33\u5b9a\u8fbe\u523099%\u4ee5\u4e0a\u7684\u6b63\u786e\u7387\u3002 \u76ee\u524d\u80fd\u8fbe\u523099.486%\u6b63\u786e\u6027\u7684\u6a21\u578b\u662f\uff1a\u628a\u5df2\u6709\u768442k\u6570\u636e\u6309\u716780%, 20%\u7684\u6bd4\u4f8b\u5206\u6210training,testing\u6570\u636e\uff0c\u7136\u540e\u7528Lenet\uff08120 convolution layers [kernel = 5] + ReLU + MaxPooling + 200 convolution layer [kernel = 3] + ReLU + MaxPooling + InnerProduct [param = 200] + InnerProduct [param=10]), \u8fed\u4ee360,000\u6b21\u3002\u7136\u540e\u8fd9\u4e2a\u8fc7\u7a0b\u91cd\u590d15\u6b21\uff0c\u7136\u540e\u9009\u591a\u6570Ensemble\u3002 \u6700\u540e\u653e\u51e0\u4e2a\u795e\u7ecf\u7f51\u7edc\u8bc6\u522b\u8d77\u6765\u8d39\u52b2\u7684\u56fe\u6848\u4f9b\u6b23\u8d4f\uff0c\u8fd8\u6709\u4e00\u4e2a\u4ecb\u7ecd\u6027\u8d28\u7684Slides\u3002 &nbsp; [&hellip;]<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[15],"tags":[139,126,140,16,141],"class_list":["post-680","post","type-post","status-publish","format-standard","hentry","category-statistics","tag-caffe","tag-deep-learning","tag-keras","tag-machine-learning","tag-mnist"],"_links":{"self":[{"href":"https:\/\/zhanxw.com\/blog\/wp-json\/wp\/v2\/posts\/680","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/zhanxw.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/zhanxw.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/zhanxw.com\/blog\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/zhanxw.com\/blog\/wp-json\/wp\/v2\/comments?post=680"}],"version-history":[{"count":0,"href":"https:\/\/zhanxw.com\/blog\/wp-json\/wp\/v2\/posts\/680\/revisions"}],"wp:attachment":[{"href":"https:\/\/zhanxw.com\/blog\/wp-json\/wp\/v2\/media?parent=680"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/zhanxw.com\/blog\/wp-json\/wp\/v2\/categories?post=680"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/zhanxw.com\/blog\/wp-json\/wp\/v2\/tags?post=680"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}