{"id":7957,"date":"2017-01-16T01:00:00","date_gmt":"2017-01-15T17:00:00","guid":{"rendered":"http:\/\/swarma.org\/%swarma-paper%\/"},"modified":"2018-12-05T22:43:52","modified_gmt":"2018-12-05T14:43:52","slug":"lstm%e5%92%8cword-embedding%e7%9a%84%e6%b7%b1%e5%85%a5%e5%89%96%e6%9e%90%ef%bc%8c%e5%8f%8arnn-attention%e7%9a%84%e6%94%b9%e8%bf%9b-%e5%b0%8f%e5%a3%b0%e5%98%9f%e5%9b%94%e4%b8%93","status":"publish","type":"post","link":"https:\/\/swarma.org\/?p=7957","title":{"rendered":"LSTM\u548cWord Embedding\u7684\u6df1\u5165\u5256\u6790\uff0c\u53caRNN\/Attention\u7684\u6539\u8fdb | \u201c\u5c0f\u58f0\u561f\u56d4\u201d\u4e13\u680f"},"content":{"rendered":"
\n
\n
\"\u96c6\u667a\"
<\/div>\n

\u5f00\u901a\u5934\u6761\u53f7\uff1a\u96c6\u667a\u4ff1\u4e50\u90e8<\/strong><\/p>\n

\n
\n
\n
\n
<\/section>\n
<\/section>\n
\n

\u4f5c\u8005\u7b80\u4ecb<\/strong><\/span><\/p>\n<\/section>\n

<\/section>\n
<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

<\/span><\/strong><\/p>\n

\u949f\u7ff0\u5ef7 <\/span><\/strong>\u6bd5\u4e1a\u4e8e\u534e\u4e2d\u79d1\u6280\u5927\u5b66\uff0c\u73b0\u5728\u7231\u56e0\u4e92\u52a8\u79d1\u6280\u53d1\u5c55\uff08\u5317\u4eac\uff09\u6709\u9650\u516c\u53f8\u62c5\u4efb\u7b97\u6cd5\u5de5\u7a0b\u5e08\uff0c\u4ece\u4e8b NLP \u76f8\u5173\u5de5\u4f5c\uff0c\u73b0\u4e3a\u96c6\u667a\u81ea\u7136\u8bed\u8a00\u5904\u7406\u7fa4\u7fa4\u4e3b\u3002<\/span><\/span><\/h2>\n

\"\u96c6\u667a\"<\/h2>\n


<\/p>\n


<\/p>\n

\n
LSTM: A Search Space Odyssey(arXiv: 1503.04069)<\/strong><\/strong><\/span><\/section>\n<\/section>\n
<\/section>\n<\/section>\n
\n

 \u4f5c\u8005\uff1a<\/span>Klaus Greff, Rupesh Kumar Srivastava, Jan Koutnik, Bas R. Steunebrink, Jurgen Schmidhuber<\/span><\/p>\n<\/blockquote>\n

 <\/span><\/strong><\/p>\n

<\/span><\/p>\n

\n
\n
\n
\n
\n
\n
\n
\u6458\u8981<\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n

<\/span>\u81ea 1995 \u5e74\u7528\u4e8e\u5faa\u73af\u795e\u7ecf\u7f51\u7edc\u7684\u957f\u77ed\u671f\u8bb0\u5fc6\u5355\u5143\uff08Long Short Term Memory, LSTM\uff09\u51fa\u73b0\u4ee5\u6765\uff0c\u5df2\u7ecf\u6709\u591a\u79cd LSTM \u7684\u53d8\u4f53\u88ab\u63d0\u51fa\u6765\u3002\u8fd1\u5e74\u6765\uff0c\u8fd9\u4e9b\u7f51\u7edc\u6a21\u578b\u5df2\u7ecf\u6210\u4e3a\u5404\u79cd\u673a\u5668\u5b66\u4e60\u95ee\u9898\u4e2d\u7684\u6700\u5148\u8fdb\u7684\u6a21\u578b\uff0c\u8fd9\u91cd\u65b0\u5f15\u53d1\u4e86\u5927\u5bb6\u5bf9\u5178\u578b LSTM \u53ca\u5176\u53d8\u4f53\u7684\u7ed3\u6784\u4e2d\u5404\u4e2a\u8ba1\u7b97\u7ec4\u4ef6\u7684\u4f5c\u7528\u7684\u5174\u8da3\u3002\u5728\u672c\u6587\u4e2d\uff0c\u6211\u4eec\u4ee5\u8bed\u97f3\u8bc6\u522b\u3001\u624b\u5199\u8bc6\u522b\u548c\u97f3\u4e50\u5efa\u6a21\u8fd9\u4e09\u4e2a\u95ee\u9898\u4e3a\u4f8b\uff0c\u5bf9\u516b\u79cd LSTM \u7ed3\u6784\u8fdb\u884c\u4e86\u5927\u89c4\u6a21\u7684\u5206\u6790\u3002\u6211\u4eec\u4f7f\u7528\u968f\u673a\u641c\u7d22\u6765\u5206\u522b\u4f18\u5316\u516b\u79cd LSTM \u7ed3\u6784\u4e2d\u7684\u8d85\u53c2\u6570\uff0c\u5e76\u7528\u5f3a\u5927\u7684 fANOVA \u6846\u67b6\u6765\u8bc4\u4f30\u8fd9\u4e9b\u8d85\u53c2\u7684\u91cd\u8981\u6027\u3002\u6211\u4eec\u5171\u8fdb\u884c\u4e86 5400 \u4e2a\u5b9e\u9a8c\uff0c\u5e76\u4ece\u8fd9\u4e9b\u5b9e\u9a8c\u4e2d\u603b\u7ed3\u51fa\u4e00\u4e9b\u89c2\u70b9 \u2014\u2014 \u8fd9\u4e9b\u5b9e\u9a8c\u7d2f\u8ba1\u8017\u8d39 CPU \u65f6\u95f4\u7ea6 15 \u5e74\uff0c\u8fd9\u662f\u7c7b\u4f3c\u7684\u5bf9 LSTM \u7f51\u7edc\u8fdb\u884c\u7684\u7814\u7a76\u4e2d\u89c4\u6a21\u6700\u5927\u7684\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u76f8\u6bd4\u7ecf\u5178\u7684 LSTM \u7ed3\u6784\uff0c\u6211\u4eec\u5b9e\u9a8c\u4e2d\u7684\u516b\u79cd\u53d8\u4f53\u5e76\u6ca1\u6709\u5728\u4e09\u4e2a\u4efb\u52a1\u4e0a\u6709\u660e\u663e\u7684\u6539\u5584\uff0c\u540c\u65f6\u6211\u4eec\u8bc1\u660e\u9057\u5fd8\u95e8(forget gate)\u548c\u8f93\u51fa\u6fc0\u6d3b\u51fd\u6570(output activation)\u662f LSTM \u4e2d\u6700\u5173\u952e\u7684\u90e8\u5206\u3002\u6211\u4eec\u8fdb\u4e00\u6b65\u89c2\u5bdf\u5230\u6a21\u578b\u7684\u8d85\u53c2\u51e0\u4e4e\u662f\u72ec\u7acb\u7684\uff0c\u540c\u65f6\u5f97\u5230\u4e86\u4e00\u4e9b\u5bf9\u8fd9\u4e9b\u8d85\u53c2\u8fdb\u884c\u6709\u6548\u8c03\u6574\u7684\u6307\u5bfc\u6027\u610f\u89c1\u3002<\/span><\/p>\n


<\/p>\n

<\/span><\/strong><\/span><\/p>\n

<\/span><\/p>\n


<\/p>\n

\n
\n
\n
\n
\n
\n
\n
\u7b80\u8bc4
<\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n

<\/span><\/strong>\u8fd9\u7bc7\u8bba\u6587\u7528\u5927\u89c4\u6a21\u7684\u6570\u636e\u5bf9\u6bd4\u4e86\u516b\u79cd\u4e0d\u540c LSTM \u53d8\u4f53\u4e4b\u95f4\u7684\u6548\u679c\u3002\u8fd9\u516b\u79cd\u53d8\u4f53\u6709\u516d\u79cd\u662f\u53bb\u9664\u7ecf\u5178 LSTM \u4e2d\u7684\u4e00\u4e9b\u8ba1\u7b97\u7ec4\u4ef6\u5f97\u5230\u7684\uff0c\u7528\u6765\u9a8c\u8bc1\u5176\u4e2d\u4e00\u4e9b\u7ec4\u4ef6\u7684\u91cd\u8981\u6027\uff1b\u5269\u4e0b\u4e24\u79cd\u4e00\u79cd\u662f\u5c06 input gate \u548c forget gate \u5408\u5e76\u7684\u7c7b\u4f3c GRU \u7684\u7ed3\u6784\uff0c\u8fd8\u6709\u4e00\u79cd\u662f\u6240\u6709 gate \u4e4b\u95f4\u4e5f\u6709 recurrent \u8fde\u63a5\u7684 full gate recurrent \u7ed3\u6784\u3002\u5b9e\u9a8c\u5f97\u5230\u7684\u4e00\u4e9b\u4e3b\u8981\u7ed3\u8bba\u6709\uff1aforget gate \u548c output activation \u662f LSTM \u4e2d\u6700\u91cd\u8981\u7684\u90e8\u4ef6\uff1b\u5b66\u4e60\u7387\u548c\u7f51\u7edc\u5927\u5c0f\u662f LSTM \u8bad\u7ec3\u4e2d\u6bd4\u8f83\u91cd\u8981\u7684\u8d85\u53c2\uff0c\u52a8\u91cf\u9879\u51e0\u4e4e\u6ca1\u4ec0\u4e48\u6548\u679c\uff1bLSTM \u7684\u5404\u4e2a\u8d85\u53c2\u51e0\u4e4e\u662f\u4e92\u76f8\u72ec\u7acb\u7684\u3002\u5982\u679c\u4e0d\u60f3\u4e86\u89e3\u592a\u591a\u7ec6\u8282\u7684\u8ba8\u8bba\uff0c\u53ef\u4ee5\u76f4\u63a5\u770b\u8bba\u6587\u7684 Conclusion \u4e00\u8282\u3002<\/span><\/p>\n


<\/p>\n


<\/p>\n

<\/span><\/p>\n

Understanding Neural Networks Through Representation Erasure<\/strong><\/strong><\/strong>(arXiv: 1612.08220)<\/strong><\/span><\/p>\n

\n
\n
\n
<\/span><\/strong><\/strong><\/span><\/section>\n<\/section>\n


<\/section>\n<\/section>\n

\n

 \u4f5c\u8005\uff1a<\/span>Jiwei Li, Will Monroe, Dan Jurafsky<\/span><\/p>\n<\/blockquote>\n

 
<\/span><\/strong><\/p>\n

<\/span><\/p>\n

\n
\n
\n
\n
\n
\n
\n
\u6458\u8981<\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\u867d\u7136\u795e\u7ecf\u7f51\u7edc\u5df2\u7ecf\u6210\u529f\u5e94\u7528\u4e8e\u8bb8\u591a\u81ea\u7136\u8bed\u8a00\u5904\u7406\u4efb\u52a1\uff0c\u4f46\u5b83\u4eec\u5728\u53ef\u89e3\u91ca\u4e0a\u4ecd\u4e0d\u591f\u3002\u5728\u672c\u6587\u4e2d\uff0c\u6211\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u901a\u7528\u7684\u65b9\u6cd5\u5206\u6790\u548c\u89e3\u91ca\u4e86\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u7684\u51b3\u7b56\u2014\u2014\u8fd9\u79cd\u65b9\u6cd5\u901a\u8fc7\u64e6\u9664\u8f93\u5165\u8868\u793a\u7684\u67d0\u4e9b\u90e8\u5206\uff0c\u6bd4\u5982\u5c06\u8f93\u5165\u8bcd\u5411\u91cf\u7684\u67d0\u4e9b\u7ef4\u3001\u9690\u85cf\u5c42\u7684\u4e00\u4e9b\u795e\u7ecf\u5143\u6216\u8005\u8f93\u5165\u7684\u4e00\u4e9b\u8bcd\u3002\u6211\u4eec\u63d0\u51fa\u4e86\u51e0\u79cd\u65b9\u6cd5\u6765\u5206\u6790\u8fd9\u79cd\u64e6\u9664\u7684\u5f71\u54cd\uff0c\u6bd4\u5982\u6bd4\u8f83\u64e6\u9664\u524d\u540e\u6a21\u578b\u7684\u8bc4\u4f30\u7ed3\u679c\u7684\u5dee\u5f02\uff0c\u4ee5\u53ca\u4f7f\u7528\u5f3a\u5316\u5b66\u4e60\u6765\u9009\u62e9\u8981\u5220\u9664\u7684\u6700\u5c0f\u8f93\u5165\u8bcd\u96c6\u5408\uff0c\u4f7f\u7528\u4e8e\u5206\u7c7b\u7684\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u7684\u5206\u7c7b\u7ed3\u679c\u53d1\u751f\u6539\u53d8\u3002\u5728\u5bf9\u591a\u4e2a NLP \u4efb\u52a1\uff08\u5305\u62ec\u8bed\u8a00\u7279\u5f81\u5206\u7c7b\u3001\u53e5\u5b50\u60c5\u611f\u5206\u6790\u3001\u6587\u6863\u7ea7\u522b\u7684 sentiment aspect prediction\uff09\u7684\u7efc\u5408\u5206\u6790\u4e2d\uff0c\u6211\u4eec\u53d1\u73b0\u6211\u4eec\u63d0\u51fa\u7684\u65b9\u6cd5\u4e0d\u4ec5\u80fd\u63d0\u4f9b\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u51b3\u7b56\u7684\u6e05\u6670\u89e3\u91ca\uff0c\u800c\u4e14\u53ef\u4ee5\u7528\u6765\u8fdb\u884c\u9519\u8bef\u5206\u6790\u3002<\/span><\/p>\n


<\/p>\n

\"\u96c6\u667a\"<\/p>\n

<\/span><\/p>\n

<\/span><\/strong><\/span><\/p>\n


<\/p>\n

\n
\n
\n
\n
\n
\n
\n
\u7b80\u8bc4
<\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\u633a\u6709\u610f\u601d\u7684\u4e00\u7bc7\u6587\u7ae0\uff0c\u901a\u8fc7\u5206\u6790\u63ed\u793a\u4e86 Word2Vec \u548c Glove \u4ea7\u751f\u7684\u8bcd\u5411\u91cf\u4e4b\u95f4\u5b58\u5728\u4e00\u4e9b\u660e\u663e\u7684\u5dee\u5f02\uff0c\u540c\u65f6\u4e5f\u8868\u660e\u8bad\u7ec3\u8bed\u6599\u4e2d\u7684\u8bcd\u9891\u5bf9\u4ea7\u751f\u7684\u8bcd\u7684\u8868\u8fbe\u6709\u5f88\u5927\u7684\u5f71\u54cd\uff1b\u5728\u53e5\u5b50\u7ea7\u522b\u7684\u60c5\u611f\u5206\u6790\u4e0a\u7684\u5b9e\u9a8c\u8868\u660e\u60c5\u611f\u8bcd\u5bf9\u60c5\u611f\u5206\u7c7b\u7ed3\u679c\u5f71\u54cd\u663e\u8457\uff0c\u6709\u610f\u601d\u7684\u662f\u8fd8\u80fd\u627e\u51fa\u6765\u4e00\u4e9b\u4f7f\u6a21\u578b\u8bef\u5206\u7c7b\u7684\u8bcd\uff1b\u5728\u6587\u6863\u7ea7\u522b\u7684 aspect prediction \u5b9e\u9a8c\u5219\u6e05\u6670\u5730\u63ed\u793a\u51fa\u6587\u6863\u4e2d\u54ea\u90e8\u5206\u6587\u672c\u548c\u7279\u5b9a\u7684 aspect \u662f\u5f3a\u5173\u8054\u7684\u3002\u540c\u65f6\u8fd9\u4e9b\u5b9e\u9a8c\u90fd\u8868\u660e\uff0c\u53cc\u5411 LSTM \u7684\u8868\u8fbe\u80fd\u529b\u6bd4\u7ecf\u5178 LSTM \u5f3a\uff0c\u7ecf\u5178 RNN \u5219\u6700\u5f31\u3002<\/span><\/p>\n


<\/span><\/p>\n

<\/strong><\/p>\n

\n
Interactive Attention for Neural Machine Translation(arXiv: 1610.0<\/span><\/strong><\/strong>5011)<\/strong><\/strong><\/span><\/section>\n<\/section>\n


<\/p>\n

\n

 \u4f5c\u8005\uff1a<\/span>\u5b5f\u51e1\u4e1c\uff0c\u5415\u6b63\u4e1c\uff0c\u674e\u822a\uff0c\u5218\u7fa4<\/span><\/p>\n<\/blockquote>\n

<\/span>
<\/span><\/strong><\/p>\n

<\/span><\/p>\n

\n
\n
\n
\n
\n
\n
\n
\u6458\u8981<\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\u5e38\u89c1\u7684\u57fa\u4e8e\u6ce8\u610f\u529b\u7684\u795e\u7ecf\u673a\u5668\u7ffb\u8bd1(Neural Machine Translation, NMT)\u5728\u751f\u6210\u76ee\u6807\u8bed\u8a00\u65f6\u8fdb\u884c\u52a8\u6001\u5bf9\u9f50\u3002\u901a\u8fc7\u91cd\u590d\u8bfb\u53d6\u7531 encoder \u4ea7\u751f\u7684\u6e90\u8bed\u8a00\u53e5\u5b50\u7684\u8868\u793a\uff0c\u6ce8\u610f\u529b\u673a\u5236\u6781\u5927\u5730\u63d0\u9ad8\u4e86 NMT \u7684\u6548\u679c\u3002\u5728\u672c\u6587\u4e2d\uff0c\u6211\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u7684\u6ce8\u610f\u529b\u673a\u5236\uff0c\u79f0\u4e3a\u300cInteractive Attention\u300d\uff0c\u5728\u7ffb\u8bd1\u65f6\uff0cdecoder \u4e0d\u4ec5\u53bb\u8bfb\u53d6\u6e90\u8bed\u8a00\u53e5\u5b50\u7684\u8868\u793a\uff0c\u800c\u4e14\u8fd8\u4f1a\u53bb\u4fee\u6539\u8fd9\u4e9b\u8868\u793a\u3002Interactive Attention \u53ef\u4ee5\u8bb0\u5f55 decoder \u548c\u6e90\u8bed\u8a00\u53e5\u5b50\u8868\u793a\u4e4b\u95f4\u7684\u4ea4\u4e92\u8fc7\u7a0b\uff0c\u56e0\u6b64\u63d0\u9ad8\u4e86\u7ffb\u8bd1\u6027\u80fd\u3002\u5728 NIST \u4e2d\u82f1\u6587\u7ffb\u8bd1\u4efb\u52a1\u4e0a\u7684\u5b9e\u9a8c\u8868\u660e\uff0c\u6211\u4eec\u7684 Interactive Attention \u6a21\u578b\u76f8\u6bd4\u539f\u59cb\u7684\u57fa\u4e8e\u6ce8\u610f\u529b\u673a\u5236\u7684\u795e\u7ecf\u673a\u5668\u7ffb\u8bd1\u6a21\u578b\u4ee5\u53ca\u5176\u4ed6\u7684\u4e00\u4e9b\u6539\u8fdb\u6a21\u578b\uff08\u5982 Coverage Model\uff09\uff0c\u5728\u6548\u679c\u4e0a\u90fd\u6709\u5f88\u5927\u7684\u63d0\u5347\u3002\u5728\u591a\u4e2a\u6d4b\u8bd5\u96c6\u4e0a\uff0c\u4f7f\u7528\u4e86 Interactive Attention \u7684\u795e\u7ecf\u673a\u5668\u7ffb\u8bd1\u7cfb\u7edf\u7684 BLEU \u503c\u6bd4\u5f00\u6e90\u7684\u57fa\u4e8e\u6ce8\u610f\u529b\u7684\u7cfb\u7edf\u5e73\u5747\u9ad8\u51fa 4.22\uff0c\u6bd4\u5f00\u6e90\u7684\u7edf\u8ba1\u673a\u5668\u7ffb\u8bd1\u7cfb\u7edf Moses \u5e73\u5747\u9ad8\u51fa 3.94\u3002<\/span><\/p>\n


<\/p>\n


<\/p>\n

\"\u96c6\u667a\"<\/p>\n

<\/span><\/strong><\/span><\/p>\n


<\/p>\n


<\/p>\n

\n
\n
\n
\n
\n
\n
\n
\u7b80\u8bc4
<\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n

 \u6587\u4e2d\u63d0\u5230\u7684 Coverage Model \u548c\u672c\u6587\u90fd\u662f\u4e3a\u4e86\u8fdb\u4e00\u6b65\u6539\u8fdb\u73b0\u6709\u7684 NMT \u65b9\u6cd5\u3001\u51cf\u5c11 over-translation \u548c under-translation \u95ee\u9898\u7684\u5de5\u4f5c\u3002\u672c\u6587\u63d0\u51fa\u7684\u529e\u6cd5\u662f\u5bf9 attention \u673a\u5236\u7684\u4e00\u4e2a\u6539\u8fdb\uff0c\u601d\u60f3\u5f88\u7b80\u5355\uff0c\u5c31\u662f\u5c06\u7531 encoder \u4ea7\u751f\u7684\u6e90\u8bed\u8a00\u53e5\u5b50\u7684\u8868\u793a \u2014\u2014 \u4e5f\u5c31\u662f encoder \u7684 hidden state \u5e8f\u5217\u89c6\u4e3a\u4e00\u4e2a\u53ef\u8bfb\u5199\u7684 memory\uff0c\u4e0d\u4ec5\u5728 decode \u7684\u65f6\u5019\u8fdb\u884c\u52a0\u6743\u7684\u8bfb\u53d6\uff0c\u8fd8\u5728\u6bcf\u4e00\u6b65 decode \u540e\u8fdb\u884c\u52a0\u6743\u7684\u4fee\u6539\u64cd\u4f5c\u3002\u8fd9\u79cd\u601d\u60f3\u5176\u5b9e\u8fd8\u53ef\u4ee5\u8ffd\u6eaf\u5230 2014 \u5e74\u7684\u795e\u7ecf\u5143\u56fe\u7075\u673a\uff0c\u5728\u5f62\u5f0f\u4e0a\u548c 2015 \u5e74\u7684 Dynamic Memory Network \u4e5f\u975e\u5e38\u63a5\u8fd1\u3002<\/span><\/p>\n


<\/p>\n

<\/span><\/p>\n


<\/p>\n

\n
Coherent Dialogue with Attention-based Languge Models
(arXiv: 1611.06997)<\/strong><\/strong><\/span><\/section>\n<\/section>\n


<\/p>\n

\n

 \u4f5c\u8005\uff1a<\/span>Hongyuan Mei, Mohit Bansal, Matthew R. Walter<\/span><\/p>\n<\/blockquote>\n

<\/span><\/p>\n


<\/span><\/strong><\/p>\n

<\/span><\/p>\n

\n
\n
\n
\n
\n
\n
\n
\u6458\u8981<\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\u6211\u4eec\u901a\u8fc7\u914d\u5907\u52a8\u6001\u6ce8\u610f\u529b\u673a\u5236(dynamic attention mechanism)\u7684 RNN \u5bf9\u8bdd\u6a21\u578b\u6765\u5efa\u6a21\u8fde\u8d2f\u5bf9\u8bdd\u7684\u8fde\u7eed\u6027\u3002\u6211\u4eec\u7684\u300c\u6ce8\u610f\u529b-RNN\u300d\u6a21\u578b\u5728\u5bf9\u8bdd\u7ee7\u7eed\u7684\u7684\u65f6\u5019\u52a8\u6001\u5730\u589e\u52a0\u5728\u5bf9\u8bdd\u5386\u53f2\u4e0a\u7684\u6ce8\u610f\u529b\u4f5c\u7528\u8303\u56f4\uff0c\u76f8\u5bf9\u5e94\u7684 seq2seq \u6a21\u578b\u4e2d\u7684\u6807\u51c6\u6ce8\u610f\u529b\u6a21\u578b\u7684\u4f5c\u7528\u8303\u56f4\u5219\u662f\u56fa\u5b9a\u7684\u3002\u8fd9\u4f7f\u5f97\u751f\u6210\u7684\u5355\u8bcd\u80fd\u548c\u5bf9\u8bdd\u5386\u53f2\u4e0a\u4e0e\u5176\u76f8\u5173\u7684\u8bcd\u5173\u8054\u8d77\u6765\u3002\u6211\u4eec\u5728\u4e24\u4e2a\u6d41\u884c\u7684\u5bf9\u8bdd\u6570\u636e\u96c6\u2014\u2014\u5f00\u653e\u57df\u7684 MovieTriples \u6570\u636e\u96c6\u548c\u5c01\u95ed\u57df\u7684 Ubuntu \u5bf9\u8bdd\u6570\u636e\u96c6\u8bc4\u4f30\u4e86\u6211\u4eec\u7684\u6a21\u578b\uff0c\u5e76\u5728\u591a\u6837\u6027\u6307\u6807\u3001\u4eba\u5de5\u8bc4\u4ef7\u3001\u51e0\u4e2a\u6307\u6807\u4e0a\u76f8\u8f83 baseline \u548c\u5f53\u524d\u6700\u5148\u8fdb\u7684\u6a21\u578b\u4e0a\u90fd\u6709\u4e0d\u5c0f\u7684\u63d0\u5347\u3002\u6211\u4eec\u7684\u5de5\u4f5c\u8fd8\u8868\u660e\uff0c\u5728\u4f7f\u7528\u7075\u6d3b\u7684\u957f\u8ddd\u79bb\u8bb0\u5fc6\u540e\uff0c\u4e00\u4e2a\u5e26\u6709\u52a8\u6001\u6ce8\u610f\u529b\u673a\u5236\u7684\u7b80\u5355 RNN \u80fd\u6bd4\u590d\u6742\u7684\u8bb0\u5fc6\u6a21\u578b\u5982 LSTM \u548c GRU \u90fd\u8868\u73b0\u5f97\u66f4\u597d\u3002\u8fdb\u4e00\u6b65\u5730\uff0c\u6211\u4eec\u901a\u8fc7\u57fa\u4e8e\u4e3b\u9898\u6a21\u578b\u7684\u91cd\u65b0\u6392\u5e8f\u6765\u63d0\u9ad8\u5bf9\u8bdd\u7684\u8fde\u8d2f\u6027\u3002<\/span><\/p>\n


<\/p>\n


<\/p>\n

\"\u96c6\u667a\"<\/p>\n

<\/span><\/p>\n


<\/p>\n

\n
\n
\n
\n
\n
\n
\n
\u7b80\u8bc4
<\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\u4e00\u822c\u6765\u8bf4\uff0c\u6211\u4eec\u770b\u5230\u7684 attention \u673a\u5236\uff0c\u90fd\u662f\u7528\u5728 encoder-decoder \u6a21\u578b\u4e0a\u7684\uff0c\u8fd9\u7bc7\u8bba\u6587\u91cc\u7684\u5c06 attention \u673a\u5236\u7528\u5728 RNN \u4e0a\u7684\u60f3\u6cd5\uff0c\u4ee5\u53ca\u6709\u300c\u5bf9\u8bdd\u662f\u66f4\u504f\u5411\u8bed\u8a00\u5efa\u6a21\u800c\u4e0d\u662f\u673a\u5668\u7ffb\u8bd1\u7684\u5de5\u4f5c\u300d\u8fd9\u6837\u7684\u89c2\u70b9\uff0c\u611f\u89c9\u633a\u6709\u610f\u601d\u7684\u3002<\/span><\/p>\n

<\/span><\/p>\n


<\/p>\n
<\/p>\n
\n
Attention-based Memory Selection Recurrent Network for Languge Modeling<\/strong>\uff08arXiv\uff1a1611.08656\uff09<\/strong>
<\/span><\/section>\n<\/section>\n

<\/strong><\/section>\n<\/section>\n


<\/p>\n

<\/span><\/p>\n

\n

 \u4f5c\u8005\uff1a<\/span>Da-Rong Liu, Shun-Pro Chuang, Hung-yi Lee<\/span><\/span><\/p>\n<\/blockquote>\n

<\/span><\/strong><\/p>\n


<\/span><\/strong><\/p>\n

<\/span><\/p>\n

\n
\n
\n
\n
\n
\n
\n
\u6458\u8981<\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\u5faa\u73af\u795e\u7ecf\u7f51\u7edc\uff08Recurrent Neural Networks, RNN\uff09\u5df2\u7ecf\u5728\u8bed\u8a00\u5efa\u6a21\u4e0a\u53d6\u5f97\u4e86\u5de8\u5927\u7684\u6210\u529f\u3002\u7136\u800c\u7531\u4e8e RNN \u4f7f\u7528\u56fa\u5b9a\u5927\u5c0f\u7684 memory\uff0c\u4e0d\u80fd\u5b58\u50a8\u5728\u53e5\u5b50\u4e2d\u5904\u7406\u8fc7\u7684\u6240\u6709\u8bcd\u7684\u4fe1\u606f\uff0c\u56e0\u6b64\u5728\u9884\u6d4b\u4e0b\u4e00\u4e2a\u8bcd\u65f6\uff0c\u6709\u7528\u7684\u957f\u671f\u8bb0\u5fc6\u5c31\u88ab\u4e22\u5931\u4e86\u3002\u5728\u672c\u6587\u4e2d\uff0c\u6211\u4eec\u63d0\u51fa\u4e86\u57fa\u4e8e\u6ce8\u610f\u529b\u673a\u5236\u7684\u8bb0\u5fc6\u9009\u62e9 RNN(Attention-based Memory Selection RNN, AMSRN)\uff0c\u8be5\u6a21\u578b\u80fd\u67e5\u770b\u524d\u4e00\u65f6\u523b\u5b58\u50a8\u5728 memory \u7684\u4fe1\u606f\u5e76\u4ece\u4e2d\u9009\u62e9\u76f8\u5173\u4fe1\u606f\u6765\u8f85\u52a9\u751f\u6210\u8f93\u51fa\u3002\u5728 AMSRN \u4e2d\uff0c\u6ce8\u610f\u529b\u673a\u5236\u9996\u5148\u5b58\u50a8\u4e86\u76f8\u5173\u4fe1\u606f\u7684 memory\uff0c\u7136\u540e\u4ece\u4e2d\u63d0\u53d6\u4fe1\u606f\u3002\u5728\u5b9e\u9a8c\u4e2d\uff0c AMSRN \u82f1\u6587\u548c\u4e2d\u6587\u7684\u8bed\u6599\u4e0a\u90fd\u53d6\u5f97\u4e86\u6bd4 LSTM \u8bed\u8a00\u6a21\u578b\u66f4\u597d\u7684\u6548\u679c\u3002\u6b64\u5916\uff0c\u6211\u4eec\u7814\u7a76\u4e86\u71b5\u6765\u5bf9 attention \u6743\u91cd\u8fdb\u884c\u6b63\u5219\u5316\uff0c\u5e76\u7528\u4e4b\u6765\u5bf9\u6ce8\u610f\u529b\u673a\u5236\u5728\u8bed\u8a00\u6a21\u578b\u4e0a\u7684\u4f5c\u7528\u8fdb\u884c\u53ef\u89c6\u5316\u3002<\/span><\/p>\n

\"\u96c6\u667a\"<\/p>\n


<\/p>\n


<\/p>\n

\n
\n
\n
\n
\n
\n
\n
\u7b80\u8bc4
<\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n

  \u548c\u300aCoherent Dialogue with Attention-based Languge Models\u300b\u7c7b\u4f3c\u7684\u601d\u60f3\uff0c\u76f4\u63a5\u5728 RNNLM \u4e0a\u52a0\u4e0a attention\uff0c\u6a21\u578b\u90e8\u5206\u8bb2\u5f97\u5f88\u6e05\u695a\uff0c\u4e0d\u8fc7\u5185\u5bb9\u6bd4\u8f83\u5c11\u5c31\u662f\u4e86\u3002<\/span><\/p>\n


<\/p>\n


<\/p>\n


<\/p>\n

\n
SampleRNN: An Unconditional End-to-End Neural Audio Generation Model<\/strong>(arXiv: 1612.07837)<\/strong><\/strong><\/span><\/section>\n<\/section>\n


<\/p>\n

<\/section>\n

<\/span><\/p>\n

\n

 \u4f5c\u8005\uff1a<\/span>Soroush Mehri, Kundan Kumar, etc.<\/span><\/span><\/p>\n<\/blockquote>\n

<\/span><\/strong><\/p>\n

 <\/span><\/strong><\/p>\n

<\/span><\/p>\n

\n
\n
\n
\n
\n
\n
\n
\u6458\u8981<\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\u5728\u672c\u6587\u4e2d\u6211\u4eec\u63d0\u51fa\u4e86\u4e00\u4e2a\u65b0\u7684\u6a21\u578b\uff0c\u7528\u4e8e\u4e00\u6b21\u751f\u6210\u4e00\u4e2a\u97f3\u9891\u6837\u672c\u7684 unconditional \u7684\u97f3\u9891\u751f\u6210(unconditional\u7ffb\u8bd1\u4e0d\u80fd)\u3002\u6211\u4eec\u7684\u6a21\u578b\u901a\u8fc7\u5c42\u7ea7\u7ed3\u6784\u5c06\u81ea\u56de\u5f52\u7684\u591a\u5c42\u611f\u77e5\u673a\u548c RNN \u7ed3\u5408\u8d77\u6765\uff0c\u5b9e\u9a8c\u8868\u660e\u8fd9\u6837\u7684\u6a21\u578b\u5728\u4e09\u4e2a\u4e0d\u540c\u6027\u8d28\u7684\u6570\u636e\u96c6\u4e0a\uff0c\u90fd\u80fd\u591f\u6355\u83b7\u975e\u5e38\u957f\u8de8\u5ea6\u4e0a\u7684\u65f6\u95f4\u5e8f\u5217\u7684\u6f5c\u5728\u53d8\u5316\u3002\u4eba\u5de5\u8bc4\u4f30\uff0c\u8868\u660e\u6211\u4eec\u7684\u6a21\u578b\u76f8\u6bd4\u5176\u4ed6\u97f3\u9891\u751f\u6210\u6a21\u578b\u6548\u679c\u66f4\u597d\u3002\u6b64\u5916\u6211\u4eec\u8fd8\u5c55\u793a\u4e86\u6a21\u578b\u4e0d\u540c\u90e8\u5206\u5bf9\u6a21\u578b\u6574\u4f53\u6027\u80fd\u7684\u8d21\u732e\u662f\u5982\u4f55\u4ea7\u751f\u7684\u3002<\/span><\/p>\n


<\/p>\n

\"\u96c6\u667a\"<\/p>\n


<\/p>\n


<\/p>\n

\n
\n
\n
\n
\n
\n
\n
\u7b80\u8bc4
<\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n

<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\u672c\u6587\u6240\u8bf4\u7684\u5c42\u7ea7\u7ed3\u6784\u5927\u81f4\u662f\u8fd9\u6837\u7684\uff0c\u6bcf\u5c42\u90fd\u662f\u4e00\u4e2a RNN \uff0c\u4f46\u662f\u6bcf\u5c42\u8f93\u5165\u7684\u8bed\u97f3\u7279\u5f81\u7684\u5e27\u957f\u4ece\u524d\u5230\u540e\u4f9d\u6b21\u51cf\u5c0f\uff0c\u5e76\u4e14\u524d\u4e00\u5c42\u7684 hidden state \u8f93\u51fa\u4f1a\u4f5c\u4e3a\u4e0b\u4e00\u4e2a RNN \u5c42\u7684\u989d\u5916\u8f93\u5165\uff0c\u8fd9\u6837\u76f8\u5f53\u4e8e\u589e\u5f3a\u4e86 RNN \u5f97\u5230\u7684\u957f\u671f\u8bb0\u5fc6\uff0c\u56e0\u6b64\u6700\u540e\u6574\u4e2a\u6a21\u578b\u5728\u5efa\u6a21\u957f\u5ea6\u4e3a 512 \u7684\u97f3\u9891\u6837\u672c\u5e8f\u5217\u65f6\uff0c\u76f8\u6bd4\u957f\u5ea6\u4e3a 32 \u7684\u97f3\u9891\u6837\u672c\u5e8f\u5217\u4e5f\u53ea\u662f\u7a0d\u6709\u53d8\u5dee\u3002\u8fd9\u79cd\u5728\u4e0d\u540c\u5c42\u7ea7\u5bf9\u4e0d\u540c\u7c92\u5ea6\u7684\u8f93\u5165\u8fdb\u884c\u5efa\u6a21\u5e76\u4e14\u4e92\u76f8\u589e\u5f3a\u7684\u65b9\u6cd5\uff0c\u548c\u8fd1\u671f\u4e00\u4e9b\u5728\u95ee\u7b54\u4e0a\u505a\u7684\u5206\u5c42\u7684 attention \u673a\u5236\u53ef\u4ee5\u8bf4\u6709\u4e00\u4e9b\u5171\u901a\u4e4b\u5904\uff0c\u5373\u4f7f\u662f\u5bf9\u97f3\u9891\u751f\u6210\u6ca1\u4ec0\u4e48\u5174\u8da3\uff0c\u4e5f\u63a8\u8350\u7a0d\u5fae\u8bfb\u4e00\u4e0b\u3002<\/span><\/p>\n


<\/span><\/p>\n

\u67e5\u770b\u5f80\u671f\u7cbe\u5f69\u56de\u987e\uff1a
<\/span><\/p>\n

\u5173\u4e8e\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u538b\u7f29(\u4e0a) | \u201c\u5c0f\u58f0\u561f\u56d4\u201d\u4e13\u680f<\/a>
<\/span><\/p>\n

\u5173\u4e8e\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u538b\u7f29(\u4e0b) | \u201c\u5c0f\u58f0\u561f\u56d4\u201d\u4e13\u680f<\/a>
<\/span><\/p>\n


<\/strong><\/span><\/p>\n

\u60f3\u77e5\u9053\u66f4\u591a\u5173\u4e8eNLP\u65b9\u9762\u7684\u5e72\u8d27\uff1f<\/strong><\/span>
<\/span><\/p>\n

\uff08\u6216\u8005\u70b9\u51fb\u9605\u8bfb\u539f\u6587\uff09<\/span><\/strong>
<\/span><\/p>\n

\n
\n
\n
\n
\n
\n
\n
\u5fae\u4fe1\u89c2\u770b\u5730\u5740<\/span><\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n
\n
\n
<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\"\u96c6\u667a\"
<\/p>\n


<\/p>\n

\n
\n
\n
\n
<\/section>\n<\/section>\n<\/section>\n
\n
\n
\n
<\/section>\n<\/section>\n
<\/section>\n
\n
\n
<\/section>\n<\/section>\n
\n
\n
\n
\n

<\/span><\/strong>\u81ea\u7136\u8bed\u8a00\u5904\u7406\u662f\u8ba9\u8ba1\u7b97\u673a\u7406\u89e3\u4eba\u7c7b\u8bed\u8a00\u7684\u524d\u6cbf\u6280\u672f\u3002\u6df1\u5ea6\u5b66\u4e60\u9769\u547d\u7684\u7206\u53d1\u4e5f\u540c\u6837\u6df1\u523b\u5730\u5f71\u54cd\u5230\u4e86\u81ea\u7136\u8bed\u8a00\u5904\u7406\u8fd9\u4e00\u5b66\u79d1\u7684\u53d1\u5c55\u3002\u672c\u7cfb\u5217\u8bfe\u7a0b\u5c06\u5168\u9762\u4ecb\u7ecd\u81ea\u7136\u8bed\u8a00\u5904\u7406\u4e0e\u6df1\u5ea6\u5b66\u4e60\u7684\u524d\u6cbf\u77e5\u8bc6\uff0c\u662f\u4e00\u5957\u5b9e\u60e0\u4f18\u8d28\u7684\u81ea\u7136\u8bed\u8a00\u5904\u7406\u5165\u95e8\u8bfe\u7a0b\u3002\u5176\u9996\u6b21\u5f00\u5c55\u662f\u96c6\u667a\u4ff1\u4e50\u90e8\u4e3e\u529e\u7684\u4ece2016\u5e747\u6708\u523011\u6708\u7684\u7ebf\u4e0a\u7cfb\u5217\u8bfb\u4e66\u4f1a\u3002<\/span><\/strong><\/p>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n

\n
<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n<\/section>\n


<\/section>\n


<\/p>\n


<\/p>\n


<\/p>\n


<\/p>\n

<\/span><\/p>\n

\n
\n

\u96c6\u667a\u4ff1\u4e50\u90e8<\/span><\/strong><\/p>\n

\u82f1\u6587\u540d\uff1a Swarm Agents Club \uff0c<\/span>\u7b80\u79f0\u201c SA Club<\/span> \u201d\uff0c\u6210\u7acb\u4e8e 2008 \u5e74 \uff0c \u662f\u4e00\u4e2a\u4ece\u4e8b\u5b66\u672f\u7814\u7a76\u3001\u4eab\u53d7\u79d1\u5b66\u4e50\u8da3\u7684\u63a2\u7d22\u8005\u7684\u56e2\u4f53 \u3002\u5021\u5bfc\u4ee5\u5e73\u7b49\u5f00\u653e\u7684\u6001\u5ea6\u3001\u79d1\u5b66\u5b9e\u8bc1\u7684\u7cbe\u795e\uff0c\u8fdb\u884c\u8de8\u5b66\u79d1\u7684\u7814\u7a76\u4e0e\u4ea4\u6d41\uff0c\u529b\u56fe\u642d\u5efa\u4e00\u4e2a\u4e2d\u56fd\u7684 \u201c \u6ca1\u6709\u56f4\u5899\u7684\u7814\u7a76\u6240 \u201d\u3002<\/strong><\/span><\/span><\/p>\n

\u96c6\u667aQQ\u5b98\u65b9\u7fa4\uff1a292641157<\/strong><\/span><\/p>\n

\u96c6\u667a\u4ff1\u4e50\u90e8\u7684\u53e3\u53f7\uff1a<\/span><\/p>\n

\u8ba9\u82f9\u679c\u7838\u5f97\u66f4\u731b\u70c8\u4e9b\u5427\uff01\uff01\uff01<\/p>\n

\u96c6\u667a\u4ff1\u4e50\u90e8\u7684\u4f7f\u547d\uff08\u5373\u76ee\u6807\uff09\uff1a<\/span><\/p>\n

\u8425\u9020\u81ea\u7531\u4ea4\u6d41\u5b66\u672f\u601d\u60f3\u7684\u5c0f\u751f\u5883\uff0c<\/p>\n

 \u5b55\u80b2\u5f00\u521b\u6027\u7684\u79d1\u5b66\u53d1\u73b0\u3002<\/p>\n

\"\u96c6\u667a\"<\/p>\n

\u8ba9\u82f9\u679c\u7838\u5f97\u66f4\u731b\u70c8\u4e9b\u5427\uff01<\/span><\/p>\n

\n
\"\u96c6\u667a\"
<\/section>\n<\/section>\n

\u957f\u6309\u8bc6\u522b\u4e8c\u7ef4\u7801\uff0c\u5173\u6ce8\u96c6\u667aClub\uff0c<\/span><\/p>\n

\u8ba9\u6211\u4eec\u79bb\u79d1\u5b66\u63a2\u7d22\u66f4\u8fd1\u4e00\u6b65\u3002<\/span><\/p>\n<\/section>\n<\/section>\n

\n

\u59cb\u53d1\u4e8e\u5fae\u4fe1\u516c\u4f17\u53f7\uff1a \u96c6\u667a\u4ff1\u4e50\u90e8 <\/p>\n<\/blockquote>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

\u5f00\u901a\u5934\u6761\u53f7\uff1a\u96c6\u667a\u4ff1\u4e50\u90e8 \u4f5c\u8005\u7b80\u4ecb \u949f\u7ff0\u5ef7 \u6bd5\u4e1a\u4e8e\u534e\u4e2d\u79d1\u6280\u5927\u5b66\uff0c\u73b0\u5728\u7231\u56e0\u4e92\u52a8\u79d1\u6280\u53d1\u5c55\uff08\u5317\u4eac\uff09\u6709\u9650\u516c\u53f8\u62c5\u4efb\u7b97\u6cd5\u5de5\u7a0b\u5e08\uff0c\u4ece\u4e8b NLP \u76f8\u5173\u5de5\u4f5c\uff0c\u73b0\u4e3a\u96c6\u667a\u81ea\u7136\u8bed\u8a00\u5904\u7406\u7fa4\u7fa4\u4e3b\u3002 LSTM: A Search Space Odyssey(arXiv: 1503.04069)  \u4f5c\u8005\uff1aKlaus Gre…<\/p>\n","protected":false},"author":1,"featured_media":7958,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[19],"tags":[],"special":[],"_links":{"self":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/7957"}],"collection":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=7957"}],"version-history":[{"count":1,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/7957\/revisions"}],"predecessor-version":[{"id":7971,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/posts\/7957\/revisions\/7971"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=\/wp\/v2\/media\/7958"}],"wp:attachment":[{"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=7957"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=7957"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=7957"},{"taxonomy":"special","embeddable":true,"href":"https:\/\/swarma.org\/index.php?rest_route=%2Fwp%2Fv2%2Fspecial&post=7957"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}