IEEE International Joint Conference on Neural Networks (IJCNN 2019)\u5c06\u4e8e2019\u5e747\u670814\u65e5\u81f37\u670819\u65e5\uff0c\u5728\u5308\u7259\u5229\u5e03\u8fbe\u4f69\u65af\u4e3e\u884c\u3002IJCNN\u662f\u795e\u7ecf\u7f51\u7edc\u53ca\u76f8\u5173\u9886\u57df\u7684\u7814\u7a76\u4eba\u5458\u548c\u5176\u4ed6\u4e13\u4e1a\u4eba\u58eb\u7684\u9996\u8981\u56fd\u9645\u4f1a\u8bae\uff0c\u4e5f\u88abCCF\u63a8\u8350\u4e3a\u4eba\u5de5\u667a\u80fd\u65b9\u5411\u7684C\u7c7b\u4f1a\u8bae\u3002<\/p>\n
<\/p>\n
17\u7ea7\u540c\u5b66\u4ed8\u6c42\u7231\u5728\u5b9e\u9a8c\u5ba4\u96f7\u51ef\u8001\u5e08\u6307\u5bfc\u4e0b\uff0c\u5b8c\u6210\u4e00\u7bc7\u957f\u6587\u201dMulti-Task Learning with Capsule Networks\u201d\uff0c\u5e76\u4ee5\u786e\u8ba4\u88abIJCNN 2019\u5f55\u7528\uff01\u4e2d\u7a3f\u8bba\u6587\u7684\u7b80\u4ecb\u5982\u4e0b\uff1a<\/p>\n
<\/p>\n
\u8bba\u6587\u6807\u9898: Multi-Task Learning with Capsule Networks<\/p>\n
<\/p>\n
\u8bba\u6587\u4f5c\u8005: Kai Lei, Qiuai Fu, Yuzhi Liang*<\/p>\n
<\/p>\n
\u82f1\u6587\u6458\u8981: Multi-task learning is a machine learning approach learning multiple tasks jointly while exploiting commonalities and differences across tasks. A shared representation is learned by multi-task learning, and what is learned for each task can help other tasks be learned better. Most of existing multi-task learning methods adopt deep neural network as the classifier of each task. However, a deep neural network can exploit its strong curve-fitting capability to achieve high accuracy in training data even when the learned representation is not good enough. This is contradictory to the purpose of multi-task learning. In this paper, we propose a framework named multi-task capsule (MT-Capsule) which improves multi-task learning with capsule network. Capsule network is a new architecture which can intelligently model part-whole relationships to constitute viewpoint invariant knowledge and automatically extend the learned knowledge to different new scenarios. The experimental results on large real-world datasets show MT-Capsule can significantly outperform the state-of-the-art methods.<\/p>\n
<\/p>\n
\u4e2d\u6587\u7b80\u4ecb: \u591a\u4efb\u52a1\u5b66\u4e60\u662f\u4e00\u79cd\u5728\u5229\u7528\u4efb\u52a1\u4e4b\u95f4\u7684\u5171\u6027\u548c\u5dee\u5f02\u7684\u540c\u65f6\uff0c\u5171\u540c\u5b66\u4e60\u591a\u4e2a\u4efb\u52a1\u7684\u673a\u5668\u5b66\u4e60\u65b9\u6cd5\u3002\u5171\u4eab\u8868\u793a\u662f\u901a\u8fc7\u591a\u4efb\u52a1\u5b66\u4e60\u6765\u5b66\u4e60\u7684\uff0c\u4e3a\u6bcf\u4e2a\u4efb\u52a1\u5b66\u4e60\u7684\u5185\u5bb9\u53ef\u4ee5\u5e2e\u52a9\u5176\u4ed6\u4efb\u52a1\u66f4\u597d\u5730\u5b66\u4e60\u3002\u73b0\u6709\u7684\u591a\u4efb\u52a1\u5b66\u4e60\u65b9\u6cd5\u5927\u591a\u91c7\u7528\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u4f5c\u4e3a\u6bcf\u4e2a\u4efb\u52a1\u7684\u5206\u7c7b\u5668\uff0c\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u53ef\u4ee5\u5229\u7528\u5176\u5f3a\u5927\u7684\u66f2\u7ebf\u62df\u5408\u80fd\u529b\u6765\u5b9e\u73b0\u8bad\u7ec3\u6570\u636e\u7684\u9ad8\u7cbe\u5ea6\uff0c\u5373\u4f7f\u5728\u5b66\u4e60\u7684\u8868\u793a\u8fd8\u4e0d\u591f\u597d\u7684\u60c5\u51b5\u4e0b\uff0c\u7136\u800c\u8fd9\u4e0e\u591a\u4efb\u52a1\u5b66\u4e60\u7684\u76ee\u7684\u662f\u76f8\u77db\u76fe\u7684\u3002\u672c\u6587\u63d0\u51fa\u4e86\u4e00\u4e2a\u591a\u4efb\u52a1\u80f6\u56ca\u7f51\u7edc\uff08MT-Capsule\uff09\u6846\u67b6\uff0c\u8be5\u6846\u67b6\u5229\u7528\u80f6\u56ca\u7f51\u7edc\u6539\u8fdb\u4e86\u591a\u4efb\u52a1\u5b66\u4e60\u3002\u80f6\u56ca\u7f51\u7edc\u662f\u4e00\u79cd\u65b0\u578b\u7684\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\uff0c\u5b83\u53ef\u4ee5\u667a\u80fd\u5730\u5bf9\u90e8\u5206\u6574\u4f53\u5173\u7cfb\u8fdb\u884c\u5efa\u6a21\uff0c\u6784\u6210\u89c6\u70b9\u4e0d\u53d8\u7684\u77e5\u8bc6\uff0c\u5e76\u81ea\u52a8\u5c06\u6240\u5b66\u77e5\u8bc6\u6269\u5c55\u5230\u4e0d\u540c\u7684\u65b0\u573a\u666f\u4e2d\u3002\u5728\u5927\u578b\u771f\u5b9e\u6570\u636e\u96c6\u4e0a\u7684\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0cMT-Capsule\u53ef\u4ee5\u663e\u8457\u4f18\u4e8e\u76ee\u524d\u6700\u597d\u7684\u65b9\u6cd5\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"
IEEE International Joint Conference on Neural Networks […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[16,31,4],"tags":[],"_links":{"self":[{"href":"https:\/\/www.icnlab.cn\/index.php?rest_route=\/wp\/v2\/posts\/6264"}],"collection":[{"href":"https:\/\/www.icnlab.cn\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.icnlab.cn\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.icnlab.cn\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.icnlab.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=6264"}],"version-history":[{"count":1,"href":"https:\/\/www.icnlab.cn\/index.php?rest_route=\/wp\/v2\/posts\/6264\/revisions"}],"predecessor-version":[{"id":6265,"href":"https:\/\/www.icnlab.cn\/index.php?rest_route=\/wp\/v2\/posts\/6264\/revisions\/6265"}],"wp:attachment":[{"href":"https:\/\/www.icnlab.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=6264"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.icnlab.cn\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=6264"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.icnlab.cn\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=6264"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}