\u4fde\u795d\u826f<\/p>\n
\u6458 \u8981\uff1a\u8fd1\u5e74\u6765, \u4ee5\u6df1\u5ea6\u5b66\u4e60\u4e3a\u6838\u5fc3\u7684\u4eba\u5de5\u667a\u80fd\u6280\u672f, \u53d6\u5f97\u4e86\u4e00\u7cfb\u5217\u91cd\u5927\u7a81\u7834.\u672c\u6587\u5c06\u5c31\u4eba\u5de5\u667a\u80fd\u7684\u4ea7\u4e1a\u5316\u70ed\u6f6e, \u4e3b\u8981\u7814\u7a76\u6d41\u6d3e\u53ca\u53d1\u5c55\u5386\u53f2, \u4ee5\u6df1\u5ea6\u5b66\u4e60\u4e3a\u6838\u5fc3\u7684\u6210\u529f\u5e94\u7528, \u4ee5\u53ca\u5b58\u5728\u7684\u4e00\u4e9b\u95ee\u9898\u548c\u4eca\u540e\u7684\u53ef\u80fd\u7814\u7a76\u65b9\u5411\u505a\u4e00\u4e2a\u4ecb\u7ecd.<\/p>\n
\u5173\u952e\u8bcd\uff1a\u4eba\u5de5\u667a\u80fd \u6df1\u5ea6\u5b66\u4e60 \u795e\u7ecf\u7f51\u7edc <\/p>\n
Review of progress on artificial intelligence<\/p>\n
YU Zhuliang<\/p>\n
Abstract\uff1aRecently, the artificial intelligence ( AI) , especially with the deep learning techniques, has achieved great success in various applications.This paper gives an overview on the artificial intelligence, including aspects of its commercialized development, the many tribes of AI with origin and history, successful stories based on deep learning, as well as the remaining challenges and possible development trends in future.<\/p>\n
Keyword\uff1aartificial intelligence (AI) deep learning neural network <\/p>\n
0 \u5f15\u8a00 <\/p>\n
\u4ece1956\u5e74\u8fbe\u7279\u8305\u65af\u4f1a\u8bae\u9996\u6b21\u5b9a\u4e49\u201c\u4eba\u5de5\u667a\u80fd\u201d (Artificial Intelligence, AI) \u5f00\u59cb, AI\u7814\u7a76\u5df2\u7ecf\u5386\u4e86\u51e0\u6b21\u5386\u53f2\u6d6e\u6c89.\u5728\u4e00\u6b21\u53c8\u4e00\u6b21\u7684\u9ad8\u6f6e\u548c\u4f4e\u8c37\u7684\u4ea4\u66ff\u4e2d, \u4e0d\u53ef\u5426\u8ba4, AI\u65e0\u8bba\u662f\u5728\u7406\u8bba\u8fd8\u662f\u5728\u5b9e\u8df5\u4e0a\u90fd\u53d6\u5f97\u4e86\u624e\u5b9e\u7684\u8fdb\u6b65, \u4eba\u7c7b\u5bf9\u4e8e\u667a\u80fd\u7684\u7406\u89e3\u8fdb\u4e00\u6b65\u52a0\u6df1.\u5c24\u5176\u662f\u8fd1\u671f\u4ee5\u6df1\u5ea6\u5b66\u4e60 (Deep Learning, DL) \u4e3a\u4ee3\u8868\u7684AI\u6280\u672f\u53d6\u5f97\u4e86\u7a81\u7834\u6027\u7684\u8fdb\u5c55, \u4ece\u800c\u5728\u5168\u4e16\u754c\u8303\u56f4\u5185\u53c8\u6380\u8d77\u4e86\u4e00\u4e2aAI\u7814\u7a76\u70ed\u6f6e.\u4e0e\u4ee5\u5f80\u4e0d\u540c\u7684\u662f, \u8fd9\u6b21\u7684\u7814\u7a76\u70ed\u6f6e\u540c\u65f6\u4f34\u968f\u7740AI\u5546\u4e1a\u5316\u6d6a\u6f6e, \u5b9e\u9a8c\u5ba4\u6210\u679c\u5f88\u5feb\u5c31\u8fdb\u5165\u5de5\u4e1a\u754c, \u751a\u81f3\u5de5\u4e1a\u754c\u5728\u8fd9\u80a1\u70ed\u6f6e\u4e2d\u4e5f\u7ad9\u5728\u4e86\u5b66\u672f\u7814\u7a76\u7684\u524d\u6cbf, \u8fd9\u5728\u4ee5\u5f80\u7684\u6280\u672f\u53d1\u5c55\u53f2\u4e0a\u662f\u975e\u5e38\u7f55\u89c1\u7684. <\/p>\n
2015\u5e747\u6708, \u4eba\u5de5\u667a\u80fd\u88ab\u5199\u5165\u300a\u56fd\u52a1\u9662\u5173\u4e8e\u79ef\u6781\u63a8\u8fdb\u201c\u4e92\u8054\u7f51+\u201d\u884c\u52a8\u7684\u6307\u5bfc\u610f\u89c1\u300b;2016\u5e743\u6708, \u4eba\u5de5\u667a\u80fd\u4e00\u8bcd\u88ab\u5199\u5165\u201c\u5341\u4e09\u4e94\u201d\u89c4\u5212\u7eb2\u8981;2016\u5e745\u6708, \u56fd\u5bb6\u53d1\u5c55\u6539\u9769\u59d4\u5458\u4f1a\u7b49\u56db\u90e8\u95e8\u8054\u5408\u4e0b\u53d1\u300a\u201c\u4e92\u8054\u7f51+\u201d\u4eba\u5de5\u667a\u80fd\u4e09\u5e74\u884c\u52a8\u5b9e\u65bd\u65b9\u6848\u300b;\u674e\u514b\u5f3a\u603b\u7406\u7684\u653f\u5e9c\u5de5\u4f5c\u62a5\u544a\u4e2d\u4e5f\u63d0\u5230\u4e86\u4eba\u5de5\u667a\u80fd\u4ea7\u4e1a\u53d1\u5c55;\u4e2d\u56fd\u79d1\u5b66\u6280\u672f\u90e8\u201c\u79d1\u6280\u521b\u65b02030\u2014\u91cd\u5927\u9879\u76ee\u201d\u8fd1\u671f\u6216\u5c06\u65b0\u589e\u201c\u4eba\u5de5\u667a\u80fd2.0\u201d, \u4eba\u5de5\u667a\u80fd\u5c06\u8fdb\u4e00\u6b65\u4e0a\u5347\u4e3a\u56fd\u5bb6\u6218\u7565.\u8fd9\u5145\u5206\u53ef\u4ee5\u770b\u51fa\u6211\u56fd\u5bf9AI\u7684\u91cd\u89c6\u7a0b\u5ea6.2017\u5e74, \u4e2d\u56fd\u5de5\u7a0b\u9662\u9662\u520a\u4fe1\u606f\u4e0e\u7535\u5b50\u5de5\u7a0b\u5b66\u90e8\u5206\u520a\u300a\u4fe1\u606f\u4e0e\u7535\u5b50\u5de5\u7a0b\u524d\u6cbf (\u82f1\u6587) \u300b\u51fa\u7248\u4e86\u201cArtificial Intelligence 2.0\u201d\u4e13\u9898, \u6f58\u4e91\u9e64\u7b49\u591a\u4f4d\u9662\u58eb\u53ca\u4e13\u5bb6\u5b66\u8005\u5bf9AI2.0\u6240\u6d89\u53ca\u7684\u5927\u6570\u636e\u667a\u80fd\u3001\u7fa4\u4f53\u667a\u80fd\u3001\u8de8\u5a92\u4f53\u667a\u80fd\u3001\u6df7\u5408\u589e\u5f3a\u667a\u80fd\u548c\u81ea\u4e3b\u667a\u80fd\u7b49\u8fdb\u884c\u4e86\u6df1\u5ea6\u9610\u8ff0. <\/p>\n
\u9762\u5bf9\u4eba\u5de5\u667a\u80fd\u70ed\u6f6e, \u6211\u4eec\u8be5\u5982\u4f55\u7406\u89e3, \u770b\u5f85\u5176\u8fdb\u6b65?\u53c8\u5982\u4f55\u4e86\u89e3\u5176\u529f\u80fd\u548c\u9650\u5236?\u5df2\u7ecf\u6709\u4e0d\u5c11\u4e66\u7c4d[1]<\/a>\u548c\u8bba\u6587[2]<\/a>\u8ba8\u8bba\u4e86\u4e0a\u8ff0\u95ee\u9898, \u672c\u6587\u5c06\u4ece\u4eba\u5de5\u667a\u80fd\u7684\u4ea7\u4e1a\u5316\u6d6a\u6f6e\u3001\u5b66\u672f\u6d41\u6d3e\u548c\u7814\u7a76\u65b9\u6cd5, \u4ee5\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u4e3a\u4e3b\u7684AI\u53d1\u5c55\u5386\u53f2\u3001\u8fd1\u671f\u6210\u679c\u548c\u5b58\u5728\u95ee\u9898\u7b49\u8bf8\u591a\u65b9\u9762\u5bf9\u4eba\u5de5\u667a\u80fd\u505a\u4e00\u4e2a\u7684\u4ecb\u7ecd, \u5e0c\u671b\u80fd\u5bf9\u8bfb\u8005\u4e86\u89e3AI\u6709\u6240\u5e2e\u52a9. <\/p>\n \u6ce81\u8fbe\u7279\u8305\u65af\u4f1a\u8bae\u4e0a\u5b9a\u4e49\u7684\u4eba\u5de5\u667a\u80fd\u662f\u6307\u7528\u8ba1\u7b97\u673a\u6a21\u62df\u4eba\u7684\u903b\u8f91\u601d\u7ef4, \u5b9e\u9645\u4e0a\u8fd9\u4e2a\u5b9a\u4e49\u6bd4\u8f83\u9002\u5408\u57fa\u4e8e\u7b26\u53f7\u903b\u8f91\u7684\u6f14\u7ece\u7cfb\u7edf (\u7b26\u53f7\u5b66\u6d3e) , \u5982\u4e13\u5bb6\u7cfb\u7edf\u7b49.\u4f46\u4eba\u7c7b\u8fd8\u6709\u5f52\u7eb3\u603b\u7ed3\u80fd\u529b (\u8054\u7ed3\u5b66\u6d3e) .\u4e25\u683c\u6765\u8bb2, \u8fd9\u4e0d\u5305\u62ec\u5728\u72ed\u4e49\u7684\u4eba\u5de5\u667a\u80fd\u5f53\u4e2d, \u6240\u4ee5\u795e\u7ecf\u7f51\u7edc\u3001\u6a21\u7cca\u903b\u8f91\u548c\u9057\u4f20\u7b97\u6cd5\u7b49\u7ed3\u5408\u53e6\u7acb\u4e86\u201c\u8ba1\u7b97\u667a\u80fd\u201d.\u4e3a\u4e86\u4fbf\u4e8e\u8868\u8ff0, \u6211\u4eec\u6587\u4e2d\u91c7\u7528\u4e86\u5e7f\u4e49\u4e0a\u7684\u4eba\u5de5\u667a\u80fd, \u8fd9\u4e2a\u6982\u5ff5\u548c\u201c\u673a\u5668\u667a\u80fd\u201d[1]<\/a>\u4e00\u81f4, \u5305\u542b\u4e86\u4e00\u5207\u673a\u5668\u5177\u6709\u7684\u667a\u80fd. <\/p>\n 1 \u4eba\u5de5\u667a\u80fd\u5546\u4e1a\u5316\u6d6a\u6f6e<\/p>\n 20\u4e16\u7eaa\u672b, \u5f53\u4ee5\u795e\u7ecf\u7f51\u7edc\u4e3a\u4e3b\u6d41\u7684AI\u7814\u7a76\u53c8\u4e00\u6b21\u8dcc\u5165\u4f4e\u8c37\u7684\u65f6\u5019, \u52a0\u62ff\u5927\u591a\u4f26\u591a\u5927\u5b66\u7684Hinton\u6559\u6388\u7b49\u8fd8\u662f\u575a\u5b88\u9635\u5730, \u8f9b\u52e4\u8015\u8018, \u5e76\u57282006\u5e74\u83b7\u5f97\u4e86\u7a81\u7834[3-4]<\/a>.2012\u5e74\u4ed6\u548c\u4e24\u4f4d\u5b66\u751f\u6210\u7acb\u201c\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u7814\u7a76\u201d (DNN Research) \u516c\u53f8, \u6570\u4e2a\u6708\u540e\u88abGoogle\u6536\u8d2d, \u4ece\u6b64Hinton\u6559\u6388\u8eab\u517c\u591a\u4f26\u591a\u5927\u5b66\u6559\u6388\u548cGoogle\u7814\u7a76\u8005\u7684\u53cc\u91cd\u8eab\u4efd.Google\u968f\u540e\u65a5\u8d444\u4ebf\u7f8e\u5143\u6536\u8d2d\u4eba\u5de5\u667a\u80fd\u521d\u521b\u7684\u524d\u6cbf\u4eba\u5de5\u667a\u80fd\u4f01\u4e1aDeep Mind.\u53e6\u5916, Google\u8fd8\u6536\u8d2d\u4e86\u4e4c\u514b\u5170\u9762\u90e8\u8bc6\u522b\u6280\u672f\u5f00\u53d1\u5546Viewdle. <\/p>\n \u7d27\u968fHinton\u6559\u6388\u7684\u6b65\u4f10, \u7ebd\u7ea6\u5927\u5b66Yann Le Cun\u6559\u6388, 2013\u5e74\u5e95\u88ab\u8058\u8bf7\u4e3aFacebook\u4eba\u5de5\u667a\u80fd\u7814\u7a76\u9662\u7684\u603b\u7ba1;\u65af\u5766\u798f\u5927\u5b66\u5434\u6069\u8fbe (Andrew Ng) \u6559\u6388, 2014\u5e74\u88ab\u767e\u5ea6\u8058\u4efb\u4e3a\u9996\u5e2d\u79d1\u5b66\u5bb6\u8d1f\u8d23\u201c\u767e\u5ea6\u5927\u8111\u201d\u7684\u8ba1\u5212 (2017\u5e74\u5df2\u7ecf\u8f9e\u804c) ;\u65af\u5766\u798f\u5927\u5b66\u674e\u98de\u98de\u6559\u6388 (FeiFei Li) \u6210\u4e3a\u8c37\u6b4c\u4e91\u8ba1\u7b97\u90e8\u95e8\u7684\u8d1f\u8d23\u4eba\u4e4b\u4e00.\u8fd9\u4e9b\u73b0\u8c61\u4e00\u65b9\u9762\u8bf4\u660e\u4eba\u5de5\u667a\u80fd\u73b0\u5728\u53d7\u5de5\u4e1a\u754c\u7684\u6b22\u8fce\u7a0b\u5ea6, \u540c\u65f6\u4e5f\u8bf4\u660e\u4e86\u4eba\u5de5\u667a\u80fd\u76ee\u524d\u7684\u53d1\u5c55\u8d8b\u52bf\u662f\u5b66\u672f\u7814\u7a76\u548c\u4f01\u4e1a\u5f00\u53d1\u7684\u5feb\u901f\u6df1\u5ea6\u7ed3\u5408. <\/p>\n \u4e3a\u4e86\u8fce\u5408AI\u7684\u53d1\u5c55\u70ed\u6f6e, \u5927\u91cf\u7684\u5f00\u6e90\u5b66\u4e60\u5e73\u53f0\u4e0d\u65ad\u95ee\u4e16, \u8d3e\u626c\u6e05\u7684Caffe\u3001Google\u7684Tensor Flow\u3001Facebook\u7684FBLearner Flow\u3001Tesla\u9886\u8854\u7684Open AI\u4ee5\u53ca\u767e\u5ea6\u6df1\u5ea6\u673a\u5668\u5b66\u4e60\u5f00\u6e90\u5e73\u53f0\u7b49, \u90fd\u4e3aAI\u7684\u7814\u7a76\u548c\u4ea7\u4e1a\u5f00\u53d1\u8d77\u5230\u4e86\u5de8\u5927\u7684\u63a8\u52a8\u4f5c\u7528.\u5f53\u7136\u8fd8\u6709\u5f88\u591a\u5176\u4ed6\u5de5\u4e1a\u5de8\u5934\u7684\u4ea7\u54c1, \u4f8b\u5982IBM\u7684\u6c83\u68ee\u7cfb\u7edf\u3001\u5fae\u8f6f\u7684\u540c\u58f0\u7ffb\u8bd1\u7b49. <\/p>\n \u79d1\u6280\u53d1\u5c55, \u4eba\u624d\u4e3a\u672c.AI\u4e5f\u4e0d\u4f8b\u5916, \u4f01\u4e1a\u5bf9\u4e8eAI\u4eba\u624d\u7684\u62a2\u593a\u66f4\u662f\u8d8b\u4e8e\u767d\u70ed\u5316.\u76ee\u524dAI\u9886\u57df, \u5c24\u5176\u662f\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u9762\u4e34\u7f3a\u4e4f\u4e13\u5bb6\u7684\u56f0\u5883.\u7531\u4e8e\u8fd9\u4e2a\u9886\u57df\u521a\u521a\u5f00\u59cb\u53d1\u5c55, \u6240\u4ee5\u4e13\u5bb6, \u5373\u4f7f\u662f\u535a\u58eb\u6bd5\u4e1a\u751f\u90fd\u7279\u522b\u5c11.\u5434\u6069\u8fbe\u6559\u6388\u66fe\u603b\u7ed3\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u4eba\u624d\u532e\u4e4f\u7684\u51e0\u4e2a\u539f\u56e0:\u9996\u5148\u662f\u6570\u636e, \u83b7\u53d6\u89e3\u51b3\u67d0\u4e9b\u9886\u57df\u7684\u95ee\u9898\u7684\u6570\u636e\u5e38\u5e38\u975e\u5e38\u56f0\u96be;\u5176\u6b21\u662f\u8ba1\u7b97\u57fa\u7840\u548c\u67b6\u6784\u5de5\u5177, \u5305\u62ec\u8ba1\u7b97\u673a\u786c\u4ef6\u548c\u8f6f\u4ef6, \u5165\u95e8\u4e0d\u6613;\u6700\u540e\u662f\u8fd9\u4e2a\u9886\u57df\u7684\u5de5\u7a0b\u5e08\u57f9\u517b\u65f6\u95f4\u957f.\u4e3a\u4e86\u89e3\u51b3\u4e0a\u8ff0\u95ee\u9898, \u5de5\u4e1a\u754c\u7684\u79d1\u6280\u5de8\u5934, \u5982Google\u3001Facebook\u3001Twitter\u3001\u767e\u5ea6\u7b49\u7eb7\u7eb7\u901a\u8fc7\u6536\u8d2d\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u7684\u521d\u521b\u516c\u53f8\u6765\u62db\u63fd\u4eba\u624d.\u5176\u4e2d\u6700\u4e3a\u5178\u578b\u7684\u662fGoogle, \u5b83\u901a\u8fc7\u4e0d\u65ad\u6536\u8d2d\u6df1\u5ea6\u5b66\u4e60\u9886\u57df\u7684\u516c\u53f8, \u62a2\u5230\u4e00\u6279\u4e16\u754c\u4e00\u6d41\u4e13\u5bb6.\u603b\u800c\u8a00\u4e4b, \u4eba\u5de5\u667a\u80fd\u4ea7\u4e1a\u7684\u53d1\u5c55, \u4f7f\u5f97\u5176\u76f8\u5173\u9886\u57df\u7684\u4eba\u624d\u6210\u4e3a\u7a00\u7f3a\u4e4b\u5b9d, \u8fd9\u5bf9\u8be5\u9886\u57df\u7684\u7814\u7a76\u4eba\u5458\u6765\u8bb2, \u65e2\u662f\u673a\u9047, \u4e5f\u662f\u6311\u6218. <\/p>\n 2 \u4eba\u5de5\u667a\u80fd\u7684\u4e3b\u8981\u7814\u7a76\u5b66\u6d3e<\/p>\n \u4eba\u7c7b\u7684\u667a\u80fd\u4e3b\u8981\u5305\u62ec\u5f52\u7eb3\u603b\u7ed3\u548c\u903b\u8f91\u6f14\u7ece\u4e24\u5927\u7c7b.\u6211\u4eec\u5927\u91cf\u7684\u611f\u77e5\u5904\u7406, \u5982\u89c6\u542c\u89c9\u3001\u8eab\u4f53\u611f\u77e5\u5904\u7406\u7b49\u90fd\u662f\u4e0b\u610f\u8bc6\u7684, \u5c5e\u4e8e\u5f52\u7eb3\u603b\u7ed3\u667a\u80fd.\u800c\u6570\u5b66\u63a8\u5bfc\u3001\u903b\u8f91\u63a8\u7406\u7b49\u90fd\u662f\u57fa\u4e8e\u516c\u7406\u7cfb\u7edf\u7684\u7b26\u53f7\u6f14\u7ece\u65b9\u6cd5.\u7531\u4e8e\u5728\u53d1\u5c55\u8fc7\u7a0b\u4e2d, \u5bf9\u4e8e\u667a\u80fd\u7684\u7406\u89e3\u7684\u4e0d\u540c, \u6e10\u6e10\u5f62\u6210\u4e86\u51e0\u4e2a\u7ecf\u5178\u5b66\u6d3e.\u6bcf\u4e2a\u5b66\u6d3e\u4ece\u4e0d\u540c\u7684\u89d2\u5ea6\u770b\u5f85\u95ee\u9898, \u63d0\u51fa\u89e3\u51b3\u65b9\u6848.\u6bd4\u5982\u6700\u4e3a\u4e3b\u8981\u7684\u4e24\u4e2a\u5b66\u6d3e:\u7b26\u53f7\u5b66\u6d3e\u548c\u8054\u7ed3\u5b66\u6d3e, \u524d\u8005\u4ece\u54f2\u5b66\u3001\u903b\u8f91\u5b66\u548c\u5fc3\u7406\u5b66\u51fa\u53d1, \u5c06\u5b66\u4e60\u89c6\u4e3a\u9006\u5411\u6f14\u7ece, \u4f7f\u7528\u9884\u5148\u5b58\u5728\u7684\u77e5\u8bc6\u6765\u89e3\u51b3\u95ee\u9898, \u5927\u591a\u6570\u4e13\u5bb6\u7cfb\u7edf\u4f7f\u7528\u7b26\u53f7\u5b66\u6d3e\u7684\u65b9\u6cd5;\u540e\u8005\u4e13\u6ce8\u4e8e\u901a\u8fc7\u795e\u7ecf\u5143\u4e4b\u95f4\u7684\u8fde\u63a5\u6765\u63a8\u5bfc\u8868\u793a\u77e5\u8bc6, \u8be5\u5b66\u6d3e\u805a\u7126\u4e8e\u7269\u7406\u5b66\u548c\u795e\u7ecf\u79d1\u5b66, \u5e76\u76f8\u4fe1\u5927\u8111\u7684\u9006\u5411\u5de5\u7a0b, \u4ed6\u4eec\u7528\u53cd\u5411\u4f20\u64ad\u7b97\u6cd5\u6765\u8bad\u7ec3\u4eba\u5de5\u795e\u7ecf\u7f51\u7edc\u4ee5\u83b7\u53d6\u7ed3\u679c[5]<\/a>.\u5176\u4ed6\u5b66\u6d3e, \u5982\u8fdb\u5316\u5b66\u6d3e\u5728\u9057\u4f20\u5b66\u548c\u8fdb\u5316\u751f\u7269\u5b66\u7684\u57fa\u7840\u4e0a\u5f97\u51fa\u7ed3\u8bba, \u8d1d\u53f6\u65af\u5b66\u6d3e\u6ce8\u91cd\u7edf\u8ba1\u5b66\u548c\u6982\u7387\u63a8\u7406, \u7c7b\u63a8\u5b66\u6d3e\u66f4\u591a\u662f\u5173\u6ce8\u5fc3\u7406\u5b66\u548c\u6570\u5b66\u4f18\u5316\u6765\u63a8\u65ad\u76f8\u4f3c\u6027\u5224\u65ad. <\/p>\n \u867d\u7136\u4e0a\u8ff0\u4e3b\u6d41\u5b66\u6d3e\u5404\u81ea\u90fd\u53d6\u5f97\u4e86\u5f88\u5927\u7684\u6210\u5c31, \u4f46\u662f\u5176\u5404\u81ea\u91c7\u7528\u7684\u7814\u7a76\u65b9\u6cd5\u90fd\u9047\u5230\u4e86\u8bf8\u591a\u56f0\u96be, \u800c\u4e14\u8fd9\u4e9b\u5b66\u6d3e\u5bf9\u4e8eAI\u7684\u7814\u7a76\u601d\u8def\u548c\u65b9\u6cd5\u96be\u4ee5\u5f62\u6210\u4e00\u4e2a\u7edf\u4e00\u7684\u6846\u67b6.\u4e3a\u4e86\u66f4\u597d\u5730\u7406\u89e3AI\u7684\u672c\u8d28, \u672c\u6587\u62df\u91c7\u7528\u6587\u732e[1, 5]\u4e2d\u7684\u89c2\u70b9\u6765\u4ecb\u7ecd\u4eba\u5de5\u667a\u80fd\u7684\u4e3b\u6d41\u7814\u7a76\u65b9\u6cd5. <\/p>\n \u57fa\u4e8e\u5bf9\u4e8e\u201c\u673a\u5668\u667a\u80fd\u662f\u7531\u4ec0\u4e48\u51b3\u5b9a\u201d\u8fd9\u4e2a\u95ee\u9898\u7684\u56de\u7b54, \u4eba\u5de5\u667a\u80fd\u9886\u57df70\u591a\u5e74\u7684\u53d1\u5c55\u5f62\u6210\u4e863\u5927\u4e3b\u6d41\u7684\u7814\u7a76\u65b9\u6cd5:\u7ed3\u6784\u6a21\u62df\u3001\u529f\u80fd\u6a21\u62df\u548c\u884c\u4e3a\u6a21\u62df.\u8fd93\u79cd\u7814\u7a76\u65b9\u6cd5\u4e4b\u95f4\u7f3a\u4e4f\u4e00\u79cd\u5185\u5728\u7684\u8054\u7cfb, \u4f7f\u5f97\u4eba\u4eec\u666e\u904d\u8ba4\u4e3aAI\u7406\u8bba\u4e0d\u6210\u4f53\u7cfb.\u4e3a\u4e86\u89e3\u51b3\u4e0a\u8ff0\u95ee\u9898, \u6587\u732e[1, 5]\u63d0\u51fa\u4e86\u7b2c4\u7c7b\u65b9\u6cd5, \u5373\u673a\u5236\u6a21\u62df. <\/p>\n 2.1 \u7ed3\u6784\u6a21\u62df<\/p>\n \u8fd1\u4ee3\u79d1\u5b66\u5f3a\u8c03\u201c\u7ed3\u6784\u51b3\u5b9a\u8bba\u201d, \u8ba4\u4e3a\u53ea\u8981\u7cfb\u7edf\u7684\u7ed3\u6784\u6e05\u695a\u4e86, \u529f\u80fd\u4e5f\u5c31\u8ba4\u8bc6\u6e05\u695a\u4e86.\u6700\u5148\u63d0\u51fa\u6765\u7684\u667a\u80fd\u6a21\u62df\u5c31\u662f\u7ed3\u6784\u6a21\u62df\u7684\u601d\u8def, \u5176\u4ee3\u8868\u6027\u7684\u6210\u679c\u6709\u795e\u7ecf\u5143\u7684MP\u6a21\u578b[6]<\/a>\u3001\u591a\u5c42\u611f\u77e5\u5668MLP\u6a21\u578b[7]<\/a>\u548c\u4eba\u5de5\u795e\u7ecf\u7f51\u7edc[8]<\/a>\u7b49.\u5728\u673a\u68b0\u7cfb\u7edf\u7684\u7814\u7a76\u4e2d\u975e\u5e38\u6709\u7528\u7684\u7ed3\u6784\u51b3\u5b9a\u8bba, \u5728\u667a\u80fd\u7cfb\u7edf\u7684\u7814\u7a76\u4e2d\u5374\u5b58\u5728\u5f88\u591a\u95ee\u9898.\u7ed3\u6784\u53ea\u662f\u786c\u4ef6\u57fa\u7840, \u4e0d\u80fd\u5b8c\u5168\u786e\u5b9a\u7cfb\u7edf\u7684\u667a\u80fd\u884c\u4e3a.\u6b63\u5982\u73b0\u5728\u5bf9\u4e8e\u5355\u4e2a\u795e\u7ecf\u5143\u7684\u751f\u7406\u6d3b\u52a8\u673a\u5236\u5df2\u7ecf\u7814\u7a76\u5f97\u975e\u5e38\u6df1\u5165, \u4f46\u662f\u5927\u91cf\u7684\u795e\u7ecf\u5143\u8fde\u63a5\u5728\u4e00\u8d77\u6240\u5448\u73b0\u7684\u667a\u80fd\u884c\u4e3a\u76ee\u524d\u7684\u7406\u89e3\u8fd8\u662f\u975e\u5e38\u80a4\u6d45.\u4f5c\u4e3a\u7ed3\u6784\u6a21\u62df\u7684\u4e3b\u6d41\u6210\u529f\u65b9\u6cd5, \u4eba\u5de5\u795e\u7ecf\u7f51\u7edc\u5177\u6709\u975e\u5e38\u7a81\u51fa\u7684\u4f18\u70b9:1) \u4eba\u5de5\u795e\u7ecf\u7f51\u7edc\u5177\u6709\u6bd4\u8f83\u89c4\u8303\u7684\u7ed3\u6784;2) \u7cfb\u7edf\u5177\u6709\u5927\u91cf\u53ef\u4ee5\u8c03\u8282\u7684\u53c2\u6570, \u81ea\u7531\u5ea6\u5927, \u53ef\u4ee5\u5b9e\u73b0\u975e\u5e38\u591a\u6837\u7684\u7cfb\u7edf;3) \u7f51\u7edc\u5177\u6709\u5e76\u884c\u5904\u7406\u7684\u673a\u5236;4) \u4fe1\u606f\u5206\u5e03\u5b58\u50a8, \u63d0\u4f9b\u4e86\u4f18\u826f\u7684\u8bb0\u5fc6\u548c\u8054\u60f3\u80fd\u529b;5) \u7cfb\u7edf\u5177\u6709\u5f88\u5f3a\u7684\u81ea\u9002\u5e94\u80fd\u529b\u548c\u9ad8\u5ea6\u7684\u5bb9\u9519\u80fd\u529b.\u5f53\u7136\u4eba\u5de5\u795e\u7ecf\u7f51\u7edc\u4e5f\u5177\u6709\u5f88\u591a\u5f31\u70b9\u548c\u5c40\u9650\u6027:1) \u4eba\u7684\u667a\u80fd\u7cfb\u7edf\u4e0d\u4ec5\u7ed3\u6784\u590d\u6742, \u800c\u4e14\u673a\u5236\u6df1\u5965, \u4e0d\u662f\u73b0\u5728\u7684\u4eba\u5de5\u795e\u7ecf\u7f51\u7edc\u6240\u80fd\u5b8c\u5168\u6a21\u62df\u7684;2) \u4eba\u5de5\u795e\u7ecf\u7f51\u7edc\u80fd\u505a\u5230\u7684\u89c4\u6a21\u548c\u4eba\u8111\u76f8\u6bd4\u8fd8\u662f\u592a\u5c0f;3) \u795e\u7ecf\u5143\u7684\u5de5\u4f5c\u673a\u7406\u8fc7\u5ea6\u7b80\u5316, \u96be\u4ee5\u4fdd\u8bc1\u7cfb\u7edf\u80fd\u6700\u7ec8\u6a21\u62df\u4eba\u8111.\u53e6\u5916, \u4eba\u5de5\u795e\u7ecf\u7f51\u7edc\u5c06\u4e00\u5207\u667a\u80fd\u90fd\u5f52\u4e3a\u6570\u503c\u8ba1\u7b97, \u5728\u672a\u80fd\u56de\u7b54\u4efb\u4f55\u95ee\u9898\u662f\u5426\u90fd\u53ef\u4ee5\u5f62\u5f0f\u5316\u6216\u8005\u6570\u503c\u5316\u8fd9\u4e2a\u95ee\u9898\u4e4b\u524d, \u8fd9\u79cd\u6570\u503c\u5316\u8ba1\u7b97\u80fd\u5426\u5b8c\u5168\u6a21\u62df\u4eba\u7c7b\u667a\u80fd\u8fd8\u662f\u4e00\u4e2a\u95ee\u9898. <\/p>\n 2.2 \u529f\u80fd\u6a21\u62df<\/p>\n \u9762\u5bf9\u7ed3\u6784\u6a21\u62df\u6240\u5b58\u5728\u7684\u95ee\u9898, \u53e6\u5916\u4e00\u7c7b\u89c2\u70b9\u8ba4\u4e3a, \u4eba\u5de5\u667a\u80fd\u7684\u7814\u7a76\u65e0\u9700\u53bb\u7406\u4f1a\u667a\u80fd\u7684\u5177\u4f53\u7ed3\u6784, \u53ea\u8981\u80fd\u591f\u6a21\u62df\u667a\u529b\u529f\u80fd\u5373\u53ef, \u8fd9\u5c31\u662f\u201c\u529f\u80fd\u4e3b\u5bfc\u8bba\u201d\u4e0b\u7684\u529f\u80fd\u6a21\u62df\u601d\u8def.\u5b9e\u9645\u4e0a, \u529f\u80fd\u6a21\u62df\u7684\u6700\u5178\u578b\u4ee3\u8868\u5c31\u662f\u4f20\u7edf\u7684\u4eba\u5de5\u667a\u80fd, \u5982\u4e13\u5bb6\u7cfb\u7edf[9]<\/a>\u662f\u5176\u6700\u6210\u529f\u7684\u5e94\u7528.\u529f\u80fd\u6a21\u62df\u4e5f\u5e38\u5e38\u88ab\u79f0\u4e3a\u7b26\u53f7\u4e3b\u4e49\u3001\u903b\u8f91\u4e3b\u4e49\u3001\u5fc3\u7406\u5b66\u6d3e.\u7b26\u53f7\u4e3b\u4e49\u8ba4\u4e3a\u4eba\u5de5\u667a\u80fd\u6e90\u4e8e\u6570\u7406\u903b\u8f91.\u6570\u7406\u903b\u8f91\u572820\u4e16\u7eaa30\u5e74\u4ee3\u5f00\u59cb\u5e94\u7528\u4e8e\u63cf\u8ff0\u667a\u80fd\u884c\u4e3a, \u5e76\u5728\u8ba1\u7b97\u673a\u4e0a\u5b9e\u73b0\u903b\u8f91\u6f14\u7ece\u7cfb\u7edf.\u540e\u6765\u7b26\u53f7\u4e3b\u4e49\u8005\u8fdb\u4e00\u6b65\u53d1\u5c55\u4e3a\u542f\u53d1\u5f0f\u7b97\u6cd5\u2014\u4e13\u5bb6\u7cfb\u7edf\u2014\u77e5\u8bc6\u5de5\u7a0b\u7406\u8bba\u548c\u6280\u672f.\u8fd9\u65b9\u9762\u7684\u7814\u7a76\u4e00\u5f00\u59cb\u53d6\u5f97\u4e86\u4e0d\u5c11\u6210\u7ee9, \u4f46\u662f\u4e00\u76f4\u88ab\u6279\u8bc4\u4e3a\u96be\u4ee5\u89e3\u51b3\u5b9e\u9645\u95ee\u9898.\u76f4\u5230\u4e13\u5bb6\u7cfb\u7edf\u51fa\u73b0, \u4e3a\u5de5\u4e1a\u3001\u7ecf\u6d4e\u548c\u793e\u4f1a\u9886\u57df\u5e26\u6765\u4e86\u6210\u529f\u7684\u65b9\u6848, \u5982\u7b2c\u4e00\u4e2a\u4e13\u5bb6\u7cfb\u7edfDENDRAL[9]<\/a>\u7528\u4e8e\u8d28\u8c31\u4eea\u5206\u6790\u6709\u673a\u5316\u5408\u7269\u7684\u5206\u5b50\u7ed3\u6784, MYCIN\u533b\u7597\u4e13\u5bb6\u7cfb\u7edf[10]<\/a>\u7528\u4e8e\u6297\u751f\u7d20\u836f\u7269\u6cbb\u7597\u7b49.\u7b26\u53f7\u4e3b\u4e49\u66fe\u7ecf\u5728\u4eba\u5de5\u667a\u80fd\u9886\u57df\u4e2d\u4e00\u679d\u72ec\u79c0, \u4e3a\u4eba\u5de5\u667a\u80fd\u53d1\u5c55\u505a\u51fa\u4e86\u6781\u5927\u8d21\u732e, \u6211\u56fd\u5434\u6587\u4fca\u9662\u58eb\u5173\u4e8e\u51e0\u4f55\u5b9a\u7406\u673a\u68b0\u5316\u8bc1\u660e\u5c31\u662f\u5176\u4e2d\u4e00\u9879\u975e\u5e38\u91cd\u8981\u7684\u6210\u679c.\u5f53\u7136, \u529f\u80fd\u6a21\u62df\u4e5f\u5177\u6709\u660e\u663e\u7684\u7f3a\u70b9:\u7cfb\u7edf\u7684\u667a\u80fd\u6c34\u5e73\u4e0e\u83b7\u5f97\u7684\u77e5\u8bc6\u6c34\u5e73\u6709\u5f88\u5927\u7684\u5173\u7cfb.\u800c\u4e14\u5f88\u591a\u77e5\u8bc6\u83b7\u53d6\u56f0\u96be, \u4e13\u5bb6\u77e5\u8bc6\u5145\u6ee1\u77db\u76fe\u548c\u504f\u9762, \u518d\u52a0\u4e0a\u73b0\u6709\u903b\u8f91\u7406\u8bba\u7684\u5c40\u9650\u6027\u7b49, \u4f7f\u5f97\u529f\u80fd\u6a21\u62df\u5728\u53d1\u5c55\u8fc7\u7a0b\u4e2d\u4e5f\u56f0\u96be\u91cd\u91cd. <\/p>\n 2.3 \u884c\u4e3a\u6a21\u62df<\/p>\n \u5728\u529f\u80fd\u6a21\u62df\u548c\u7ed3\u6784\u6a21\u62df\u90fd\u66b4\u9732\u4e86\u5404\u81ea\u7684\u7f3a\u9677\u540e, 20\u4e16\u7eaa90\u5e74\u4ee3, \u5f00\u59cb\u51fa\u73b0\u4e86\u884c\u4e3a\u6a21\u62df\u7684\u601d\u8def, \u5373\u201c\u884c\u4e3a\u8868\u73b0\u8bba\u201d.\u8be5\u89c2\u70b9\u8ba4\u4e3a, \u65e0\u8bba\u91c7\u7528\u4ec0\u4e48\u6837\u7684\u7ed3\u6784\u548c\u5177\u6709\u4ec0\u4e48\u6837\u7684\u529f\u80fd, \u53ea\u8981\u7cfb\u7edf\u80fd\u8868\u73b0\u51fa\u667a\u80fd\u884c\u4e3a (\u5728\u5916\u754c\u523a\u6fc0\u65f6\u80fd\u591f\u4ea7\u751f\u6070\u5f53\u7684\u884c\u4e3a\u54cd\u5e94) , \u5c31\u7b49\u4e8e\u6a21\u62df\u4e86\u667a\u80fd\u7cfb\u7edf.\u5728\u8fd9\u4e2a\u7814\u7a76\u65b9\u6cd5\u4e2d, \u9996\u5148\u662f\u673a\u5668\u611f\u77e5, \u7136\u540e\u9488\u5bf9\u611f\u77e5\u4fe1\u606f\u505a\u6a21\u5f0f\u5206\u7c7b, \u6700\u540e\u662f\u5bf9\u611f\u77e5\u5224\u65ad\u7ed3\u679c\u505a\u51fa\u7684\u6a21\u62df\u667a\u80fd\u7684\u884c\u4e3a, \u8fd9\u4e5f\u88ab\u79f0\u4e3a\u611f\u77e5-\u52a8\u4f5c\u7cfb\u7edf.\u8fd9\u65b9\u9762\u7684\u5178\u578b\u5e94\u7528\u662fBrooks\u5b8c\u6210\u7684\u6a21\u62df\u516d\u811a\u866b\u7684\u722c\u884c\u673a\u5668\u4eba\u7b49\u7cfb\u7edf[11]<\/a>.\u611f\u77e5-\u52a8\u4f5c\u7cfb\u7edf\u6d89\u53ca\u5230\u4e00\u4e2a\u91cd\u8981\u95ee\u9898:\u968f\u7740\u4efb\u52a1\u73af\u5883\u7684\u53d8\u5316, \u7cfb\u7edf\u5982\u4f55\u80fd\u81ea\u4e3b\u5b66\u4e60\u5e76\u6269\u5145\u4ece\u611f\u77e5\u5230\u52a8\u4f5c\u4e4b\u95f4\u7684\u6620\u5c04\u77e5\u8bc6?\u8fd9\u5f53\u7136\u5c31\u662f\u4e00\u4e2a\u673a\u5668\u5b66\u4e60\u95ee\u9898.\u884c\u4e3a\u6a21\u62df\u4e5f\u5177\u6709\u660e\u663e\u7684\u7f3a\u70b9:\u53ea\u6709\u90a3\u4e9b\u80fd\u7528\u884c\u4e3a\u8868\u73b0\u7684\u667a\u80fd\u624d\u80fd\u88ab\u6a21\u62df, \u53ef\u662f\u5f88\u591a\u667a\u80fd\u8fc7\u7a0b\u65e0\u6cd5\u7528\u884c\u4e3a\u76f4\u63a5\u8868\u8fbe. <\/p>\n 2.4 \u673a\u5236\u6a21\u62df<\/p>\n \u7ed3\u6784\u6a21\u62df\u3001\u529f\u80fd\u6a21\u62df\u548c\u884c\u4e3a\u6a21\u62df\u90fd\u5177\u6709\u5148\u5929\u4e0d\u8db3, \u800c\u4e14\u8fd93\u5927\u65b9\u6cd5\u4e4b\u95f4\u7f3a\u4e4f\u7406\u8bba\u4e0a\u7684\u7edf\u4e00\u6027.\u540e\u6765\u7684\u7814\u7a76\u53d1\u73b0, \u667a\u80fd\u7684\u751f\u6210\u673a\u5236\u624d\u662f\u667a\u80fd\u7cfb\u7edf\u7684\u6838\u5fc3.\u673a\u5236\u6a21\u62df\u65b9\u6cd5\u8ba4\u4e3a, \u65e0\u8bba\u5bf9\u4ec0\u4e48\u95ee\u9898\u3001\u73af\u5883\u548c\u76ee\u6807, \u667a\u80fd\u7cfb\u7edf\u7684\u751f\u6210\u673a\u5236\u5fc5\u7136\u8981\u83b7\u5f97\u201c\u95ee\u9898\u3001\u7ea6\u675f\u6761\u4ef6\u3001\u9884\u8bbe\u76ee\u6807\u201d\u7b49\u4fe1\u606f, \u7136\u540e\u63d0\u53d6\u548c\u5efa\u7acb\u76f8\u5173\u77e5\u8bc6, \u8fdb\u800c\u5728\u76ee\u6807\u63a7\u5236\u4e0b, \u5229\u7528\u4e0a\u8ff0\u4fe1\u606f\u548c\u77e5\u8bc6\u6f14\u7ece\u51fa\u6c42\u89e3\u95ee\u9898\u7684\u7b56\u7565, \u5e76\u8f6c\u5316\u4e3a\u76f8\u5e94\u7684\u667a\u80fd\u884c\u4e3a\u4f5c\u7528\u4e8e\u95ee\u9898, \u5e76\u89e3\u51b3\u95ee\u9898[1,5]<\/a>.\u8fd9\u53ef\u4ee5\u6982\u62ec\u4e3a\u4e00\u79cd\u201c\u4fe1\u606f\u2014\u77e5\u8bc6\u2014\u667a\u80fd\u8f6c\u6362\u8fc7\u7a0b\u201d.\u6839\u636e\u8fd9\u4e2a\u89c2\u70b9, \u7ed3\u6784\u6a21\u62df\u53ef\u4ee5\u8ba4\u4e3a\u662f\u201c\u4fe1\u606f\u2014\u7ecf\u9a8c\u77e5\u8bc6\u2014\u7ecf\u9a8c\u7b56\u7565\u8f6c\u6362\u8fc7\u7a0b\u201d, \u529f\u80fd\u6a21\u62df\u662f\u201c\u4fe1\u606f\u2014\u89c4\u8303\u77e5\u8bc6\u2014\u89c4\u8303\u7b56\u7565\u8f6c\u6362\u8fc7\u7a0b\u201d, \u884c\u4e3a\u6a21\u62df\u53ef\u4ee5\u8ba4\u4e3a\u662f\u201c\u4fe1\u606f\u2014\u5e38\u8bc6\u77e5\u8bc6\u2014\u5e38\u8bc6\u7b56\u7565\u8f6c\u6362\u8fc7\u7a0b\u201d.\u56e0\u6b64, \u7ed3\u6784\u6a21\u62df\u3001\u529f\u80fd\u6a21\u62df\u548c\u884c\u4e3a\u6a21\u62df\u4e09\u8005\u90fd\u662f\u5e73\u884c\u7684, \u800c\u673a\u5236\u6a21\u62df\u548c\u8c10\u5730\u7edf\u4e00\u4e86\u4e0a\u8ff03\u79cd\u6a21\u62df\u65b9\u6cd5, \u6210\u4e3a\u4e86\u4e00\u4e2a\u7edf\u4e00\u7684\u7406\u8bba. <\/p>\n \u5728AI\u53d1\u5c55\u8fc7\u7a0b\u4e2d, \u4e0a\u8ff0\u591a\u4e2a\u65b9\u6cd5\u5404\u81ea\u90fd\u51fa\u73b0\u8fc7\u81ea\u5df1\u7684\u53d1\u5c55\u5dc5\u5cf0\u548c\u4f4e\u8c37\u65f6\u671f.\u76ee\u524d\u7684AI\u70ed\u6f6e\u5219\u6e90\u4e8e\u7ed3\u6784\u6a21\u62df\u65b9\u6cd5\u65b9\u9762\u7684\u7a81\u7834, \u5373\u7531\u4e8e\u89e3\u51b3\u4e86\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u7684\u8bad\u7ec3\u95ee\u9898, \u52a0\u4e0a\u5927\u6570\u636e\u7684\u9ad8\u6027\u80fd\u8ba1\u7b97\u5e73\u53f0 (\u4e91\u8ba1\u7b97\u3001GPU\u7b49) \u53d8\u6210\u73b0\u5b9e, \u4f7f\u5f97\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u7684\u8868\u8fbe\u80fd\u529b\u5f97\u5230\u4e86\u5145\u5206\u7684\u53d1\u6325, \u5bf9AI\u7684\u53d1\u5c55\u8d77\u5230\u4e86\u63a8\u6ce2\u52a9\u6f9c\u7684\u4f5c\u7528.\u672c\u6587\u5c06\u8fdb\u4e00\u6b65\u4ee5\u6df1\u5ea6\u5b66\u4e60\u4e3a\u4e3b\u4ecb\u7ecd\u5176\u53d1\u5c55\u548c\u6210\u529f\u6848\u4f8b. <\/p>\n 3 \u795e\u7ecf\u7f51\u7edc\u53d1\u5c55\u8fc7\u7a0b\u53ca\u6df1\u5ea6\u5b66\u4e60\u7684\u5174\u8d77<\/p>\n \u8ba9\u673a\u5668\u5177\u6709\u667a\u80fd, \u662f\u4eba\u7c7b\u4e00\u76f4\u7684\u68a6\u60f3, \u4f46\u662f\u5b9e\u8d28\u6027\u7684\u8fdb\u5c55\u5374\u662f\u572820\u4e16\u7eaa50\u5e74\u4ee3\u5f00\u59cb\u7684.\u5728Mc Culloch\u548cPitts\u7684\u795e\u7ecf\u5143\u8ba1\u7b97\u6a21\u578b\u5de5\u4f5c[6]<\/a>\u57fa\u7840\u4e0a, \u5eb7\u5948\u5c14\u5927\u5b66Rosenblatt\u63d0\u51fa\u4e86\u611f\u77e5\u5668 (Perceptron) \u6a21\u578b[7]<\/a>.\u5728\u611f\u77e5\u5668\u7814\u7a76\u4e2d, Rosenblatt\u53d7\u5230Hebb\u5de5\u4f5c\u7684\u542f\u53d1, \u63d0\u51fa\u4e00\u5957\u7b97\u6cd5\u6765\u7cbe\u786e\u5b9a\u4e49\u7f51\u7edc\u7684\u5b66\u4e60\u89c4\u5219, \u8fd9\u4e5f\u662f\u9996\u4e2a\u5177\u6709\u81ea\u7ec4\u7ec7\u81ea\u5b66\u4e60\u80fd\u529b\u7684\u6a21\u578b.Hebb\u8ba4\u4e3a\u77e5\u8bc6\u548c\u5b66\u4e60\u5728\u5927\u8111\u4e2d\u4e3b\u8981\u662f\u901a\u8fc7\u795e\u7ecf\u5143\u95f4\u7a81\u89e6\u7684\u5f62\u6210\u4e0e\u53d8\u5316\u6765\u5b9e\u73b0\u7684.\u611f\u77e5\u5668\u901a\u8fc7\u8c03\u6574\u9488\u5bf9\u8f93\u5165\u503c\u7684\u6743\u91cd, \u5229\u7528\u4e00\u4e2a\u975e\u5e38\u7b80\u5355\u76f4\u89c2\u7684\u5b66\u4e60\u65b9\u6cd5, \u4ece\u8f93\u5165\u6570\u636e\u4e0a\u5b9e\u73b0\u5b66\u4e60\u529f\u80fd.Rosenblatt\u8fd8\u7528\u5b9a\u5236\u786c\u4ef6\u7684\u65b9\u6cd5\u5b9e\u73b0\u4e86\u611f\u77e5\u5668, \u5c55\u793a\u51fa\u5b83\u53ef\u4ee5\u7528\u6765\u5b66\u4e60\u5e76\u5bf920\u00d720\u50cf\u7d20\u8f93\u5165\u4e2d\u7684\u7b80\u5355\u5f62\u72b6\u8fdb\u884c\u6b63\u786e\u5206\u7c7b.\u81ea\u6b64, \u673a\u5668\u5b66\u4e60\u95ee\u4e16\u4e86.\u611f\u77e5\u5668\u4e0d\u4ec5\u662f\u65e5\u540e\u8bb8\u591a\u65b0\u7684\u795e\u7ecf\u7f51\u7edc\u6a21\u578b\u7684\u59cb\u7956, \u8fd8\u4e3a20\u4e16\u7eaa60\u5e74\u4ee3\u5e26\u6765\u4eba\u5de5\u667a\u80fd\u7684\u7b2c\u4e00\u4e2a\u70ed\u6f6e.\u8fd9\u80a1\u70ed\u6f6e\u540c\u65f6\u8fce\u6765\u4e86\u6fc0\u70c8\u7684\u6279\u8bc4.1969\u5e74, Minsky\u548cPapert\u5728\u540d\u4e3a\u300a\u611f\u77e5\u5668\u300b\u7684\u4e66\u4e2d\u63d0\u51fa\u4e86\u5f3a\u70c8\u7684\u6279\u5224.\u4ed6\u4eec\u8ba4\u4e3a\u5355\u5c42\u7684\u611f\u77e5\u5668\u7f51\u7edc\u65e0\u6cd5\u89e3\u51b3\u975e\u7ebf\u6027\u53ef\u5206\u95ee\u9898 (\u5982\u5f02\u6216\u95e8\u3001XOR\u95ee\u9898) .\u53e6\u5916, \u7f51\u7edc\u6a21\u578b\u6240\u9700\u7684\u8ba1\u7b97\u91cf\u4e5f\u8d85\u51fa\u4e86\u5f53\u65f6\u8ba1\u7b97\u673a\u7684\u80fd\u529b.\u5b66\u672f\u754c\u666e\u904d\u8ba4\u4e3a\u8fd9\u672c\u4e66\u5bf9\u4eba\u5de5\u667a\u80fd\u6b65\u5165\u7b2c\u4e00\u4e2a\u51ac\u5929\u8d77\u5230\u4e86\u63a8\u6ce2\u52a9\u6f9c\u7684\u4f5c\u7528\u2014\u2014\u2014\u4eba\u5de5\u667a\u80fd\u8fdb\u5165\u6ce1\u6cab\u5e7b\u706d\u671f, \u76f8\u5173\u8d44\u52a9\u548c\u51fa\u7248\u90fd\u906d\u51bb\u7ed3. <\/p>\n Minsky\u548cPaper\u5173\u4e8e\u611f\u77e5\u5668\u7684\u5206\u6790\u8bc1\u660e\u7528\u5355\u4e2a\u611f\u77e5\u5668\u65e0\u6cd5\u89e3\u51b3XOR\u95ee\u9898, \u6307\u51fa\u5fc5\u9700\u8981\u591a\u5c42\u611f\u77e5\u5668\u7f51\u7edc (\u6240\u8c13\u7684\u591a\u5c42\u795e\u7ecf\u7f51\u7edc) \u624d\u53ef\u4ee5\u5b8c\u6210\u4efb\u52a1, \u800c\u4e14Rosenblatt\u7684\u5b66\u4e60\u7b97\u6cd5\u5bf9\u591a\u5c42\u7f51\u7edc\u65e0\u7528.\u8fd9\u4e2a\u95ee\u9898\u6700\u7ec8\u7684\u89e3\u51b3\u65b9\u6848\u5c31\u662f\u8457\u540d\u7684\u53cd\u5411\u4f20\u64ad\u7b97\u6cd5[12-13]<\/a>, \u8be5\u65b9\u6cd5\u8ba9\u7814\u7a76\u8005\u5e7f\u6cdb\u7406\u89e3\u4e86\u5e94\u8be5\u5982\u4f55\u8bad\u7ec3\u591a\u5c42\u795e\u7ecf\u7f51\u7edc\u6765\u89e3\u51b3\u590d\u6742\u5b66\u4e60\u95ee\u9898, \u5176\u4e2d\u5305\u62ec\u975e\u7ebf\u6027\u53ef\u5206\u95ee\u9898.\u901a\u8fc7\u5728\u795e\u7ecf\u7f51\u7edc\u91cc\u589e\u52a0\u4e00\u4e2a\u6216\u8005\u591a\u4e2a\u9690\u5c42, \u53ef\u4ee5\u4f7f\u5f97\u591a\u5c42\u795e\u7ecf\u7f51\u7edc\u5177\u6709\u975e\u5e38\u5f3a\u7684\u89e3\u51b3\u590d\u6742\u95ee\u9898\u7684\u80fd\u529b.\u66f4\u6709\u8da3\u7684\u53d1\u73b0\u662f\u6570\u5b66\u80fd\u8bc1\u660e\u591a\u5c42\u524d\u9988\u795e\u7ecf\u7f51\u7edc\u662f\u666e\u9002\u6a21\u62df\u5668 (Universal Approximator) .\u672c\u8d28\u4e0a, \u591a\u5c42\u7ed3\u6784\u4f7f\u5f97\u795e\u7ecf\u7f51\u7edc\u80fd\u591f\u5728\u7406\u8bba\u4e0a\u6267\u884c\u4efb\u4f55\u51fd\u6570\u8868\u8fbe, \u5f53\u7136\u5305\u62ecXOR (\u5f02\u6216) \u95ee\u9898[14-15]<\/a>.Minsky\u63d0\u51fa\u7684\u8ba1\u7b97\u91cf\u95ee\u9898\u4e5f\u5f88\u5feb\u5f97\u5230\u4e86\u89e3\u51b3, \u4f20\u7edf\u7684\u611f\u77e5\u5668\u7528\u6240\u8c13\u201c\u68af\u5ea6\u4e0b\u964d\u201d\u7684\u7b97\u6cd5\u7ea0\u9519\u65f6, \u5176\u8fd0\u7b97\u91cf\u548c\u795e\u7ecf\u5143\u6570\u76ee\u7684\u5e73\u65b9\u6210\u6b63\u6bd4, \u5728Rumelhart\u548cHinton\u7b49[13]<\/a>\u5408\u4f5c\u7684\u8bba\u6587\u4e2d, \u7cfb\u7edf\u5730\u63d0\u51fa\u4e86\u5e94\u7528\u53cd\u5411\u4f20\u64ad\u7b97\u6cd5, \u628a\u7ea0\u9519\u7684\u8fd0\u7b97\u91cf\u4e0b\u964d\u5230\u53ea\u548c\u795e\u7ecf\u5143\u6570\u76ee\u6210\u6b63\u6bd4.Hinton\u548c\u5176\u535a\u58eb\u540eYann Le Cun\u4e8e1989\u5e74\u91c7\u7528\u7f8e\u56fd\u90ae\u653f\u7cfb\u7edf\u63d0\u4f9b\u7684\u8fd1\u4e07\u4e2a\u624b\u5199\u6570\u5b57\u7684\u6837\u672c\u6765\u8bad\u7ec3\u795e\u7ecf\u7f51\u7edc\u7cfb\u7edf, \u5728\u72ec\u7acb\u7684\u6d4b\u8bd5\u6837\u672c\u4e2d\u9519\u8bef\u7387\u4f4e\u81f35%, \u8fbe\u5230\u5b9e\u7528\u6c34\u51c6[16]<\/a>.\u968f\u540eYann Le Cun\u7b49\u8fdb\u4e00\u6b65\u8fd0\u7528\u201c\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u201d (Convoluted Neural Networks) \u7684\u6280\u672f[17]<\/a>, \u5f00\u53d1\u51fa\u5546\u4e1a\u8f6f\u4ef6, \u7528\u4e8e\u8bfb\u53d6\u94f6\u884c\u652f\u7968\u4e0a\u7684\u624b\u5199\u6570\u5b57, \u83b7\u5f97\u4e86\u5de8\u5927\u7684\u6210\u529f.\u795e\u7ecf\u7f51\u7edc\u6380\u8d77\u4e86\u7b2c\u4e8c\u6b21\u70ed\u6f6e. <\/p>\n \u4f46\u662f\u5f88\u5feb\u7814\u7a76\u8005\u53d1\u73b0\u4e86\u53cd\u5411\u4f20\u64ad\u7b97\u6cd5\u5177\u6709\u672c\u8d28\u7f3a\u9677\u2014\u2014\u2014\u68af\u5ea6\u6d88\u5931 (Vanishing Gradient Problem) [18]<\/a>, \u4e5f\u5c31\u662f\u8bf4, \u795e\u7ecf\u7f51\u7edc\u7684\u4ee3\u4ef7\u51fd\u6570 (Cost Function) \u7684\u8bef\u5dee\u4ece\u8f93\u51fa\u5c42\u5411\u8f93\u5165\u5c42\u53cd\u5411\u4f20\u64ad\u65f6, \u68af\u5ea6\u8870\u51cf\u6781\u5feb, \u5b66\u4e60\u901f\u5ea6\u53d8\u5f97\u6781\u6162, \u751a\u81f3\u65e0\u6cd5\u5b66\u4e60, \u795e\u7ecf\u7f51\u7edc\u5f88\u5bb9\u6613\u505c\u6ede\u4e8e\u5c40\u90e8\u6700\u4f18\u89e3.\u8fd9\u4f7f\u5f97\u7406\u8bba\u4e0a\u53ef\u4ee5\u5b66\u4e60\u4efb\u610f\u51fd\u6570\u7684\u591a\u5c42\u795e\u7ecf\u7f51\u7edc\u5728\u5b9e\u7528\u4e2d\u65e0\u6cd5\u5b9e\u73b0.\u540c\u65f6, \u591a\u5c42\u7f51\u7edc\u7531\u4e8e\u5177\u6709\u8f83\u591a\u7684\u53c2\u6570, \u5b66\u4e60\u81ea\u7531\u5ea6\u5927, \u7b97\u6cd5\u8bad\u7ec3\u65f6\u4f1a\u51fa\u73b0\u8fc7\u5ea6\u62df\u5408 (Overfitting) \u95ee\u9898, \u4f7f\u5f97\u5b66\u4e60\u8fc7\u7a0b\u4e2d\u8868\u73b0\u826f\u597d\u7684\u7f51\u7edc\u7684\u6cdb\u5316\u8bef\u5dee\u5f88\u5927, \u65e0\u6cd5\u771f\u6b63\u5e94\u7528.\u5728\u591a\u5c42\u7f51\u7edc\u906d\u9047\u4e0a\u8ff0\u91cd\u5927\u95ee\u9898\u7684\u65f6\u5019, \u8d1d\u5c14\u5b9e\u9a8c\u5ba4\u7684Vapnik\u63d0\u51fa\u4e86\u652f\u6301\u5411\u91cf\u673a (Support Vector Machine, SVM) \u7684\u7b97\u6cd5[19]<\/a>, \u901a\u8fc7\u4f7f\u7528\u6240\u8c13\u201c\u6838\u673a\u5236\u201d (Kernel Trick) \u7684\u975e\u7ebf\u6027\u6620\u5c04, \u4f7f\u5f97\u672c\u6765\u7ebf\u6027\u4e0d\u53ef\u5206\u7684\u6837\u672c\u6620\u5c04\u5230\u7ebf\u6027\u53ef\u5206\u7684\u9ad8\u7ef4\u7279\u5f81\u7a7a\u95f4\u4e2d.\u4ece20\u4e16\u7eaa90\u5e74\u4ee3\u521d\u5f00\u59cb, SVM\u5728\u56fe\u50cf\u548c\u8bed\u97f3\u8bc6\u522b\u4e0a\u627e\u5230\u4e86\u5e7f\u6cdb\u7684\u5e94\u7528.\u7531\u4e8e\u5176\u7406\u8bba\u5b8c\u5907\u3001\u673a\u7406\u7b80\u5355\u53d7\u5230\u4e86\u7814\u7a76\u8005\u7684\u8ffd\u6367.\u5728\u591a\u65b9\u9762\u7684\u4f5c\u7528\u4e0b, \u795e\u7ecf\u7f51\u7edc\u7684\u7814\u7a76\u518d\u4e00\u6b21\u8fdb\u5165\u4e86\u51ac\u5929. <\/p>\n \u4ece\u4e0a\u8ff0\u5206\u6790\u53ef\u4ee5\u770b\u51fa, \u795e\u7ecf\u7f51\u7edc\u7814\u7a76\u518d\u4e00\u6b21\u8fdb\u5165\u51ac\u5929\u6709\u5176\u672c\u8d28\u6027\u7684\u95ee\u9898, \u5982\u68af\u5ea6\u6d88\u5931\u3001\u8fc7\u62df\u5408\u548c\u8ba1\u7b97\u91cf\u5927\u7b49.\u5728\u8fd9\u4e2a\u51ac\u5929\u4e2d, \u6709\u4e9b\u7814\u7a76\u8005\u5982Hinton\u7b49\u4f9d\u7136\u76f8\u4fe1\u95ee\u9898\u80fd\u83b7\u5f97\u89e3\u51b3, \u5e76\u575a\u5b88\u8fd9\u5757\u9635\u5730.\u7ecf\u8fc7\u591a\u5e74\u7684\u52aa\u529b, \u6700\u7ec8\u8fce\u6765\u4e86\u795e\u7ecf\u7f51\u7edc\u7684\u590d\u5174, \u5e76\u4e14\u5f00\u542f\u4e86\u6df1\u5ea6\u5b66\u4e60\u7684\u5927\u95e8.\u5728\u795e\u7ecf\u7f51\u7edc\u7684\u8bad\u7ec3\u4e0a, \u65e0\u76d1\u7763\u5b66\u4e60\u88ab\u5f15\u5165\u5230\u4e86\u7f51\u7edc\u7684\u521d\u59cb\u5316\u4e2d.\u5728\u6587\u732e[3]\u4e2d, Hinton\u7b49\u5229\u7528\u9650\u5236\u73bb\u5c14\u5179\u66fc\u673a (RBM) \u5bf9\u795e\u7ecf\u7f51\u7edc\u5b9e\u73b0\u4e86\u65e0\u76d1\u7763\u8bad\u7ec3 (Unsupervised Training) .RBM\u4ece\u8f93\u5165\u6570\u636e\u4e2d\u901a\u8fc7\u65e0\u76d1\u7763\u8bad\u7ec3\u53d1\u73b0\u91cd\u8981\u7279\u5f81, \u5bf9\u795e\u7ecf\u7f51\u7edc\u7684\u6743\u91cd\u8fdb\u884c\u6709\u6548\u7684\u521d\u59cb\u5316, \u7136\u540e\u5c06\u591a\u5c42RBM\u53e0\u52a0\u5728\u4e00\u8d77\u5f62\u6210\u6df1\u5ea6\u7f51\u7edc, \u518d\u5bf9\u6574\u4f53\u7f51\u7edc\u7528\u53cd\u5411\u4f20\u64ad\u7b97\u6cd5\u8fdb\u884c\u5fae\u8c03, \u53d6\u5f97\u4e86\u5f88\u597d\u7684\u6548\u679c.\u53e6\u5916\u5728\u6587\u732e[20]\u4e2d\u63d0\u51fa\u7528\u4e00\u79cd\u201c\u4fee\u6b63\u7ebf\u6027\u5355\u5143\u201d (REctified Linear Unit, RELU) \u7684\u8f6c\u6362\u51fd\u6570\u6765\u66ff\u4ee3\u4f20\u7edf\u795e\u7ecf\u5355\u5143\u7684\u975e\u7ebf\u6027\u8f6c\u6362\u51fd\u6570.RELU\u51fd\u6570\u7b80\u5355, \u800c\u4e14\u5176\u5bfc\u6570\u4e3a\u5e38\u6570, \u8f93\u5165\u5c0f\u4e8e\u96f6\u65f6\u4e3a0, \u5927\u4e8e\u96f6\u65f6\u4e3a1, \u4e0d\u5b58\u5728\u4f20\u7edf\u8f6c\u6362\u51fd\u6570\u5728\u53cd\u5411\u4f20\u64ad\u8ba1\u7b97\u4e2d\u7684\u68af\u5ea6\u6d88\u5931\u95ee\u9898.\u57fa\u4e8e\u65e0\u76d1\u7763\u5b66\u4e60\u7684\u521d\u59cb\u5316\u548cRELU\u51fd\u6570\u7684\u91c7\u7528, \u5f88\u597d\u5730\u89e3\u51b3\u4e86\u591a\u5c42\u7f51\u7edc\u7684\u8bad\u7ec3\u95ee\u9898. <\/p>\n \u4e3a\u4e86\u89e3\u51b3\u591a\u5c42\u7f51\u7edc\u8bad\u7ec3\u4e2d\u7684\u8fc7\u62df\u5408\u95ee\u9898, Hinton\u7b49\u5728\u6587\u732e[21]\u4e2d\u63d0\u51fa\u4e00\u79cd\u79f0\u4e3a\u4e22\u5f03 (Dropout) \u7684\u7b97\u6cd5, \u5728\u6bcf\u6b21\u8bad\u7ec3\u4e2d, \u4ee5\u4e00\u5b9a\u7684\u6bd4\u4f8b, \u5728\u8bad\u7ec3\u4e2d\u5ffd\u7565\u8fd9\u90e8\u5206\u795e\u7ecf\u5143.\u8be5\u7b97\u6cd5\u53ef\u4ee5\u4f7f\u5f97\u7f51\u7edc\u53d8\u5f97\u66f4\u9c81\u68d2, \u907f\u514d\u8fc7\u5ea6\u62df\u5408.\u53e6\u5916, \u7531\u4e8e\u5927\u6570\u636e\u53d8\u6210\u4e86\u73b0\u5b9e, \u5c06\u5927\u6570\u636e\u7528\u4e8e\u8bad\u7ec3, \u4e5f\u662f\u89e3\u51b3\u8fc7\u62df\u5408\u7684\u4e00\u4e2a\u91cd\u8981\u7684\u9014\u5f84.\u4f8b\u5982\u73b0\u5728\u7684Image Net\u6570\u636e\u5e93, \u5c31\u53ef\u4ee5\u63d0\u4f9b\u9ad8\u8fbe\u5343\u4e07\u5f20\u56fe\u7247\u7528\u4e8e\u8bad\u7ec3.\u5927\u6570\u636e\u6210\u4e86\u6df1\u5ea6\u5b66\u4e60\u53d1\u5c55\u7684\u4e00\u4e2a\u91cd\u8981\u57fa\u7840. <\/p>\n \u5f53\u591a\u5c42\u795e\u7ecf\u7f51\u7edc\u7684\u8bad\u7ec3\u548c\u8fc7\u62df\u5408\u5f97\u5230\u4e86\u89e3\u51b3\u540e, \u7406\u8bba\u4e0a\u6211\u4eec\u53ef\u4ee5\u5229\u7528\u975e\u5e38\u6df1\u7684\u7f51\u7edc\u6765\u5b66\u4e60\u5b9e\u73b0\u4efb\u610f\u590d\u6742\u7684\u51fd\u6570.\u73b0\u5728\u7684\u6df1\u5ea6\u7f51\u7edc\u5f80\u5f80\u5177\u6709\u767e\u4e07\u5230\u6570\u4ebf\u7684\u5b66\u4e60\u53c2\u6570, \u800c\u4e14\u4e3a\u4e86\u5b9e\u73b0\u8fd9\u6837\u7684\u7f51\u7edc\u8bad\u7ec3, \u5f80\u5f80\u9700\u8981\u6d77\u91cf\u7684\u6570\u636e, \u8ba1\u7b97\u91cf\u81ea\u7136\u66f4\u6210\u4e3a\u4e86\u4e00\u4e2a\u95ee\u9898.\u5e78\u8fd0\u7684\u662f, \u968f\u7740\u4e91\u8ba1\u7b97\u7684\u666e\u53ca, \u5927\u91cf\u7684\u8ba1\u7b97\u8d44\u6e90\u53ef\u4ee5\u88ab\u4f7f\u7528.\u800c\u4e14\u65af\u5766\u798f\u5927\u5b66\u7684\u5434\u6069\u8fbe\u7b49[22]<\/a>, \u901a\u8fc7\u7528GPU\u5b9e\u73b0\u5927\u89c4\u6a21\u5b66\u4e60, \u5b9e\u73b0\u4e86\u6570\u5341\u500d\u5230\u6570\u767e\u500d\u7684\u901f\u5ea6\u63d0\u5347.\u4ece\u6b64\u4eba\u7c7b\u8fce\u6765\u4e86\u795e\u7ecf\u7f51\u7edc\u7684\u5927\u590d\u5174, \u5e76\u4e14\u6253\u5f00\u4e86\u6df1\u5ea6\u5b66\u4e60\u7684\u5927\u95e8.\u5173\u4e8e\u6df1\u5ea6\u5b66\u4e60\u66f4\u5b8c\u6574\u7684\u8bba\u6587\u8d44\u6599\u8bf7\u53c2\u8003\u6587\u732e[23]. <\/p>\n \u5982\u679c\u8ba4\u4e3a\u6df1\u5ea6\u5b66\u4e60\u7684\u63d0\u51fa\u4ec5\u4ec5\u662f\u7531\u4e8e\u7b97\u6cd5\u7684\u6539\u8fdb\u548c\u8ba1\u7b97\u80fd\u529b\u7684\u589e\u5f3a, \u90a3\u662f\u6211\u4eec\u5bf9\u6df1\u5ea6\u5b66\u4e60\u7684\u672c\u8d28\u8fd8\u8ba4\u8bc6\u5730\u4e0d\u591f\u6df1\u523b.\u5b9e\u9645\u4e0a, \u6df1\u5ea6\u5b66\u4e60\u7684\u63d0\u51fa\u548c\u73b0\u5728\u53d6\u5f97\u7684\u7a81\u98de\u731b\u8fdb\u7684\u8fdb\u5c55, \u6709\u5176\u66f4\u52a0\u91cd\u8981\u800c\u6df1\u523b\u7684\u601d\u60f3\u53d8\u5316.\u5728\u8111\u79d1\u5b66\u7814\u7a76\u4e2d\u53d1\u73b0, \u5927\u8111\u5177\u6709\u4e0d\u540c\u7684\u529f\u80fd\u533a\u57df\u4e13\u95e8\u8d1f\u8d23\u540c\u4e00\u7c7b\u4efb\u52a1, \u4f8b\u5982\u89c6\u89c9\u56fe\u50cf\u8bc6\u522b\u3001\u8bed\u97f3\u4fe1\u53f7\u5904\u7406\u548c\u6587\u5b57\u5904\u7406\u7b49.\u56e0\u6b64\u79d1\u5b66\u5bb6\u4e3a\u4e0d\u540c\u7684\u4efb\u52a1\u5f00\u53d1\u4e0d\u540c\u7684\u7b97\u6cd5, \u5982Gabor\u6ee4\u6ce2\u5668\u3001SIFT\u7279\u5f81\u63d0\u53d6\u7b97\u5b50\u3001\u9a6c\u5c14\u79d1\u592b\u968f\u673a\u573a\u7b49\u6765\u63d0\u53d6\u4fe1\u53f7\u7684\u7279\u5f81, \u6700\u7ec8\u7528\u4e8e\u6a21\u4eff\u5927\u8111\u529f\u80fd.\u4f46\u6587\u732e[24-25]\u4e2d\u7684\u7814\u7a76\u8868\u660e, \u5927\u8111\u5b9e\u9645\u4e0a\u662f\u4e00\u53f0\u901a\u7528\u5b66\u4e60\u673a\u5668 (Universal Learning Machine) , \u540c\u6837\u7684\u5b66\u4e60\u673a\u5236\u53ef\u4ee5\u7528\u4e8e\u5b8c\u5168\u4e0d\u540c\u7684\u4efb\u52a1, \u4e0d\u540c\u7684\u8111\u529f\u80fd\u533a\u53ef\u4ee5\u8f6c\u6362, \u800c\u4e14\u8f6c\u6362\u8fc7\u7a0b\u4e2d\u80fd\u81ea\u52a8\u5b66\u4e60\u7279\u5f81, \u5927\u8111\u7684\u795e\u7ecf\u7f51\u7edc\u5177\u6709\u6781\u5f3a\u7684\u53ef\u5851\u6027.\u5927\u8111\u5b66\u4e60\u7b97\u6cd5\u7684\u666e\u9002\u6027\u548c\u53ef\u5851\u6027\u4e00\u76f4\u6fc0\u52b1\u7740\u8ba1\u7b97\u673a\u79d1\u5b66\u5bb6\u4e0d\u61c8\u5730\u52aa\u529b\u63a2\u7d22.\u5386\u53f2\u6027\u7684\u7a81\u7834\u53d1\u751f\u57282006\u5e74, Hinton\u7b49[4]<\/a>\u7a81\u7834\u6df1\u5ea6\u5b66\u4e60\u7684\u6280\u672f\u74f6\u9888, \u8fdb\u800c\u5f15\u9886\u6df1\u5ea6\u5b66\u4e60\u7684\u70ed\u6f6e.\u66f4\u6709\u610f\u601d\u7684\u662f\u901a\u8fc7\u6df1\u5ea6\u5b66\u4e60\u83b7\u5f97\u7684\u7279\u5f81\u5c45\u7136\u548c\u5927\u8111\u7684\u89c6\u89c9\u5904\u7406\u8fc7\u7a0b\u975e\u5e38\u76f8\u4f3c, \u4ece\u800c\u8bc1\u660e\u6df1\u5ea6\u5b66\u4e60\u4ece\u67d0\u4e2a\u65b9\u9762\u5df2\u7ecf\u5bf9\u5927\u8111\u7684\u5b66\u4e60\u673a\u5236\u505a\u4e86\u5f88\u597d\u7684\u6a21\u4eff, \u4ece\u6570\u636e\u4e2d\u5b66\u4e60\u5230\u529f\u80fd[26]<\/a>. <\/p>\n 4 \u6df1\u5ea6\u5b66\u4e60\u4fc3\u8fdb\u7684\u4eba\u5de5\u667a\u80fd\u53d1\u5c55<\/p>\n \u6df1\u5ea6\u5b66\u4e60\u63d0\u51fa\u81f3\u4eca, \u5df2\u7ecf\u5728\u5404\u7c7b\u5e94\u7528\u4e0a\u53d6\u5f97\u4e86\u5de8\u5927\u7684\u8fdb\u5c55.\u5c24\u5176\u662f\u57fa\u4e8e\u6df1\u5ea6\u5b66\u4e60\u7684Alpha Go\u7cfb\u7edf\u4e00\u4e3e\u51fb\u8d25\u97e9\u56fd\u56f4\u68cb\u624b\u674e\u4e16\u77f3\u4ee5\u540e, \u4eba\u4eec\u90fd\u5bf9\u4ee5\u6df1\u5ea6\u5b66\u4e60\u4e3a\u4e3b\u7684AI\u7814\u7a76\u5145\u6ee1\u4e86\u671f\u5f85.\u4e8b\u5b9e\u4e0a, \u73b0\u5728\u51e0\u4e4e\u6bcf\u5929\u90fd\u53ef\u4ee5\u770b\u5230AI\u53d6\u5f97\u5404\u7c7b\u7a81\u7834\u7684\u62a5\u9053.\u611f\u5174\u8da3\u7684\u8bfb\u8005\u53ef\u4ee5\u5728\u7f51\u4e0a\u627e\u5230\u5404\u79cd\u6700\u65b0\u7684\u8fdb\u5c55\u62a5\u9053.\u672c\u6587\u5c06\u9009\u62e9\u5176\u4e2d\u51e0\u4e2a\u4e3b\u8981\u7684\u8fdb\u5c55\u8fdb\u884c\u7b80\u5355\u4ecb\u7ecd. <\/p>\n \u6df1\u5ea6\u5b66\u4e60\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\u7684\u6210\u529f\u6848\u4f8b\u9996\u63a8\u56fe\u50cf\u8bc6\u522b.2009\u5e74, \u666e\u6797\u65af\u987f\u5927\u5b66\u5efa\u7acb\u4e86\u7b2c\u4e00\u4e2a\u8d85\u5927\u578b\u56fe\u50cf\u6570\u636e\u5e93\u4f9b\u8ba1\u7b97\u673a\u89c6\u89c9\u7814\u7a76\u8005\u4f7f\u7528[27]<\/a>, \u968f\u540e\u5728\u4ee5Image Net\u4e3a\u57fa\u7840\u7684\u5927\u578b\u56fe\u50cf\u8bc6\u522b\u7ade\u8d5b\u201cImage Net Large Scale Visual Recognition Challenge 2012\u201d\u4e2d, Hinton\u56e2\u961f\u5c06\u6df1\u5ea6\u5b66\u4e60\u5e94\u7528\u5230Image Net\u56fe\u50cf\u8bc6\u522b\u95ee\u9898\u4e0a, \u6b63\u786e\u7387\u7a33\u5c45\u7b2c\u4e00, \u5e76\u4e14\u6027\u80fd\u9065\u9065\u9886\u5148\u7b2c\u4e8c\u540d\u56e2\u961f.\u8fd9\u6807\u5fd7\u7740\u6df1\u5ea6\u5b66\u4e60\u5728\u56fe\u50cf\u8bc6\u522b\u9886\u57df\u5927\u5e45\u5ea6\u8d85\u8d8a\u5176\u4ed6\u6280\u672f, \u6210\u4e3aAI\u6280\u672f\u7a81\u7834\u70b9.\u968f\u540e\u4ee5\u6df1\u5ea6\u5b66\u4e60\u4e3a\u4e3b\u7684\u56fe\u50cf\u5206\u6790\u5904\u7406\u65b9\u6cd5\u5c42\u51fa\u4e0d\u7a77, \u5982\u6df1\u5ea6\u6b8b\u4f59\u5b66\u4e60 (Deep Residual Learning) \u65b9\u6cd5\u7b49.\u76ee\u524d\u57fa\u4e8e\u6df1\u5ea6\u5b66\u4e60\u7684\u56fe\u50cf\u8bc6\u522b\u529f\u80fd\u5df2\u7ecf\u8d85\u8d8a\u4e86\u4eba\u7c7b. <\/p>\n \u53e6\u5916, Kaggle\u7f51\u7ad9\u4e3e\u529e\u4e86\u4e00\u573a\u5728\u536b\u661f\u56fe\u50cf\u4e0a\u8fdb\u884c\u573a\u666f\u7279\u5f81\u68c0\u6d4b\u7684\u6bd4\u8d5b, \u6570\u636e\u96c6\u7531\u82f1\u56fd\u56fd\u9632\u79d1\u5b66\u4e0e\u6280\u672f\u5b9e\u9a8c\u5ba4 (DSTL) \u63d0\u4f9b.\u536b\u661f\u9886\u57df\u4ea7\u751f\u7684\u5927\u91cf\u7684\u56fe\u50cf\u6570\u636e, \u975e\u5e38\u9002\u5408\u7528\u6df1\u5ea6\u5b66\u4e60\u65b9\u6cd5\u53bb\u5904\u7406, \u800c\u4e14\u6700\u7ec8\u7684\u7ade\u8d5b\u7ed3\u679c\u53d1\u73b0, \u6df1\u5ea6\u5b66\u4e60\u65b9\u6cd5\u53d6\u5f97\u4e86\u975e\u5e38\u4f18\u79c0\u7684\u6027\u80fd.\u5728\u56fe\u50cf\u7406\u89e3\u65b9\u6cd5\u4e0a, \u6df1\u5ea6\u5b66\u4e60\u65b9\u6cd5\u7ed3\u5408\u9012\u5f52\u795e\u7ecf\u7f51\u7edc (RNN) , \u5c31\u53ef\u4ee5\u5b9e\u73b0\u5bf9\u56fe\u50cf\u7684\u81ea\u7136\u8bed\u8a00\u5c42\u9762\u7684\u7406\u89e3.\u65af\u5766\u798f\u5927\u5b66\u674e\u98de\u98de\u56e2\u961f\u7ed3\u5408\u4e86\u5377\u79ef\u7f51\u7edc\u548c\u9012\u5f52\u7f51\u7edc\u5b9e\u73b0\u4e86\u56fe\u7247\u6807\u9898\u7684\u81ea\u52a8\u751f\u6210. <\/p>\n \u8bed\u8a00\u662f\u4eba\u673a\u4ea4\u6d41\u7684\u4e00\u79cd\u91cd\u8981\u9014\u5f84, \u653b\u514b\u8bed\u97f3\u8bc6\u522b\u662fAI\u5fc5\u987b\u9762\u5bf9\u7684\u95ee\u9898\u4e4b\u4e00.\u6700\u5148\u5f00\u59cb\u5728\u8bed\u97f3\u8bc6\u522b\u4e0a\u53d6\u5f97\u6210\u529f\u7684\u6df1\u5ea6\u5b66\u4e60\u65b9\u6cd5\u662fHinton\u7b49\u5728\u6587\u732e[28]\u4e2d\u7684\u65b9\u6cd5, \u8be5\u65b9\u6cd5\u7528RBM\u5bf9\u795e\u7ecf\u7f51\u7edc\u8fdb\u884c\u9884\u8bad\u7ec3, \u518d\u7528\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u6a21\u578b (DNN) \u8bc6\u522b\u8bed\u97f3.\u5728Google\u7684\u4e00\u4e2a\u57fa\u51c6\u6d4b\u8bd5\u4e2d, \u5355\u8bcd\u9519\u8bef\u7387\u964d\u4f4e\u523012.3%.\u6587\u732e[29]\u7528RNN\/LSTM\u7b49\u6280\u672f\u5728\u97f3\u4f4d\u9519\u8bef\u7387\u6d4b\u8bd5\u4e2d\u4f18\u4e8e\u540c\u671f\u7684\u6240\u6709\u5176\u4ed6\u6280\u672f.AI\u5728\u8bed\u97f3\u8bc6\u522b\u4e0a\u7684\u6210\u529f\u662f\u7ee7\u56fe\u50cf\u8bc6\u522b\u4e4b\u540e\u7684\u53c8\u4e00\u4e2a\u6280\u672f\u7a81\u7834\u70b9. <\/p>\n \u75be\u75c5\u8bca\u65ad, \u4e00\u76f4\u662f\u533b\u751f\u7684\u4e13\u5229, \u4e5f\u662f\u4e00\u4e2a\u795e\u79d8\u7684\u4e13\u4e1a\u9886\u57df.\u73b0\u5728\u901a\u8fc7\u6df1\u5ea6\u5b66\u4e60, \u6211\u4eec\u53ef\u4ee5\u8ba9\u673a\u5668\u5230\u8fbe\u533b\u5b66\u4e13\u5bb6\u7684\u8bca\u65ad\u6c34\u5e73.\u6587\u732e[30]\u4e2d\u62a5\u9053\u4e86\u5229\u7528\u6df1\u5ea6\u5b66\u4e60\u8bca\u65ad\u76ae\u80a4\u764c\u7684\u5de5\u4f5c.\u6211\u4eec\u77e5\u9053\u76ae\u80a4\u764c\u662f\u4eba\u7c7b\u6700\u5e38\u89c1\u7684\u6076\u6027\u80bf\u7624, \u76ee\u524d\u4e3b\u8981\u662f\u901a\u8fc7\u89c6\u89c9\u8bca\u65ad\u7684.\u8be5\u6587\u4f7f\u7528\u6df1\u5ea6\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u8fdb\u884c\u76ae\u80a4\u75c5\u53d8\u5206\u7c7b, \u8be5\u7f51\u7edc\u4ec5\u4f7f\u7528\u56fe\u50cf\u548c\u75be\u75c5\u6807\u7b7e\u4f5c\u4e3a\u8f93\u5165, \u5b66\u4f1a\u6b63\u786e\u5206\u7c7b.\u4ed6\u4eec\u5728\u4e24\u4e2a\u4e8c\u5206\u7c7b\u4efb\u52a1:\u89d2\u8d28\u5f62\u6210\u7ec6\u80de\u764c (Keratinocyte Carcinomas) \u548c\u826f\u6027\u8102\u6ea2\u6027\u89d2\u5316\u75c5 (Benign Seborrheic Keratoses) \u3001\u6076\u6027\u9ed1\u8272\u7d20\u7624\u548c\u666e\u901a\u7684\u75e3\u4e0a\u8fdb\u884c\u4e86\u6d4b\u8bd5, \u53d1\u73b0\u6df1\u5ea6\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u5728\u8fd9\u4e24\u4e2a\u4efb\u52a1\u4e0a\u7684\u8868\u73b0\u90fd\u8fbe\u5230\u4e86\u6240\u6709\u6d4b\u8bd5\u7684\u4e13\u5bb6\u7684\u6c34\u5e73, \u8fd9\u8bc1\u660e\u4e86\u5728\u76ae\u80a4\u764c\u8bca\u65ad\u95ee\u9898\u4e0a, \u673a\u5668\u8fbe\u5230\u4e86\u76ae\u80a4\u79d1\u4e13\u4e1a\u533b\u751f\u7684\u6c34\u5e73.\u6700\u65b0\u7684\u6587\u732e[31]\u4e2d\u62a5\u9053, \u79d1\u5b66\u5bb6\u4f7f\u7528\u80fd\u591f\u81ea\u5b66\u4e60\u7684AI\u6280\u672f, \u8ba9\u8ba1\u7b97\u673a\u5728\u9884\u6d4b\u5fc3\u810f\u75c5\u7684\u53d1\u4f5c\u4e0a\u51fb\u8d25\u4e86\u4eba\u7c7b\u533b\u751f.\u8be5\u6280\u672f\u4e00\u65e6\u6295\u5165\u4f7f\u7528, \u8fd9\u4e00\u65b0\u7684\u8bca\u7597\u624b\u6bb5\u6bcf\u5e74\u5c06\u62ef\u6551\u6570\u4ee5\u5343\u8ba1\u751a\u81f3\u767e\u4e07\u8ba1\u7684\u751f\u547d. <\/p>\n \u9664\u6b64\u4ee5\u5916, \u6df1\u5ea6\u5b66\u4e60\u5728\u6e38\u620f\u65b9\u9762\u4e5f\u53d6\u5f97\u4e86\u7a81\u7834.\u6700\u4e3a\u4e16\u4eba\u6240\u79f0\u9053\u7684\u662fAlpha Go[32]<\/a>\u5bf9\u5c40\u674e\u4e16\u77f3\u7684\u6bd4\u8d5b, Alpha Go\u4ee5\u538b\u5012\u6027\u7684\u80dc\u5229\u8d62\u4e86\u4eba\u7c7b\u9876\u7ea7\u68cb\u624b, \u5176\u4e2d\u5f88\u591a\u7cbe\u5999\u7684\u62db\u5f0f\u8ba9\u4eba\u53f9\u4e3a\u89c2\u6b62.\u5230\u5e95\u662f\u6df1\u5ea6\u5b66\u4e60\u7b97\u6cd5\u5df2\u7ecf\u50cf\u4eba\u7c7b\u4e00\u6837\u5177\u6709\u4e86\u521b\u9020\u529b\u8fd8\u662f\u795e\u7ecf\u5143\u53c2\u6570\u3001\u9002\u5f53\u7684\u7b97\u6cd5\u7ed3\u5408CPU\u7684\u8ba1\u7b97\u86ee\u529b\u4e0b\u7684\u6210\u529f, \u503c\u5f97\u4eba\u4eec\u6df1\u601d.Google\u7684Deep Mind\u56e2\u961f\u5f00\u53d1\u7684\u6df1\u5ea6Q\u7f51\u7edc (DQN) \u572849\u79cdAtari\u50cf\u7d20\u6e38\u620f\u4e2d, 29\u79cd\u8fbe\u5230\u4e43\u81f3\u8d85\u8fc7\u4eba\u7c7b\u804c\u4e1a\u9009\u624b\u7684\u6c34\u5e73[33]<\/a>. <\/p>\n \u673a\u5668\u5177\u6709\u5f3a\u5927\u7684\u8ba1\u7b97\u80fd\u529b\u3001\u5b58\u50a8\u7a7a\u95f4\u548c\u68c0\u7d22\u901f\u5ea6, \u5982\u679c\u8bf4\u673a\u5668\u5728\u903b\u8f91\u601d\u7ef4\u4e0a\u80fd\u6a21\u4eff\u5e76\u51fb\u8d25\u4eba\u7c7b\u662f\u5408\u60c5\u5408\u7406\u7684.\u90a3\u4e48\u8bf4\u673a\u5668\u80fd\u5b66\u4f1a\u827a\u672f\u9274\u8d4f, \u5219\u5c06\u4ee4\u4eba\u96be\u4ee5\u7f6e\u4fe1.\u5b9e\u9645\u4e0a, \u6587\u732e[34]\u4e2d\u62a5\u9053\u4e86\u6df1\u5ea6\u5b66\u4e60\u5df2\u7ecf\u53ef\u4ee5\u5206\u8fa8\u4e0d\u540c\u827a\u672f\u4f5c\u54c1\u7684\u5185\u5bb9\u548c\u98ce\u683c, \u5e76\u4e14\u5b66\u4f1a\u628a\u4e00\u5e45\u4f5c\u54c1\u7684\u827a\u672f\u98ce\u683c\u5e94\u7528\u5230\u5176\u4ed6\u4f5c\u54c1\u4e2d\u53bb.Google\u7684AI\u5b9e\u9a8c\u5ba4\u6210\u529f\u5b9e\u73b0\u8ba9\u673a\u5668\u5b66\u4e60\u5c0f\u8bf4\u7684\u53d9\u8ff0\u548c\u6587\u5b57\u7528\u6cd5, \u751a\u81f3\u53e5\u5b50\u7684\u6784\u9020.\u8fd9\u610f\u5473\u7740\u6df1\u5ea6\u5b66\u4e60\u53ef\u4ee5\u91cf\u5316\u5f88\u591a\u4eba\u7c7b\u72ec\u6709\u7684\u4e00\u4e9b\u6a21\u7cca\u7684\u5224\u51b3\u80fd\u529b, \u6709\u4e9b\u611f\u89c9\u4eba\u7c7b\u672c\u8eab\u4e5f\u4e0d\u6e05\u695a\u5982\u4f55\u89e3\u91ca, \u751a\u81f3\u53ea\u662f\u505c\u7559\u5728\u611f\u89c9\u5c42\u9762, \u5982\u827a\u672f\u98ce\u683c\u3001\u6587\u91c7\u7279\u8272\u4ee5\u53ca\u68cb\u7c7b\u6e38\u620f\u4e2d\u7684\u68cb\u9762\u4f18\u52bf\u7b49.\u8fd9\u4e9b\u672c\u6765\u53ea\u53ef\u8a00\u4f20, \u4e0d\u53ef\u610f\u4f1a\u7684\u6982\u5ff5\u5c45\u7136\u88ab\u673a\u5668\u5b66\u4f1a\u4e86.\u5230\u5e95\u662f\u673a\u5668\u5b66\u4e60\u5389\u5bb3, \u8fd8\u662f\u8fd9\u4e9b\u865a\u5e7b\u7684\u6982\u5ff5\u540e\u9762\u5176\u5b9e\u6709\u4e2a\u867d\u7136\u975e\u5e38\u590d\u6742, \u4f46\u662f\u660e\u786e\u53ef\u4ee5\u83b7\u5f97\u7684\u51fd\u6570\u5728\u652f\u914d, \u800c\u6df1\u5ea6\u5b66\u4e60\u53ea\u662f\u7528\u5176\u86ee\u529b\u53bb\u628a\u8fd9\u4e9b\u51fd\u6570\u53cd\u6620\u5230\u4e86\u6570\u4ebf\u51b0\u51b7\u7684\u53c2\u6570\u4e2d\u53bb? <\/p>\n 5 \u4eba\u5de5\u667a\u80fd\u53d1\u5c55\u9762\u4e34\u7684\u4e3b\u8981\u95ee\u9898\u548c\u65b0\u7684\u7814\u7a76\u65b9\u5411<\/p>\n AI\u6982\u5ff5\u63d0\u51fa\u4ee5\u540e, \u8fc5\u901f\u53d1\u5c55\u6210\u4e3a\u4e00\u95e8\u5e7f\u53d7\u5173\u6ce8\u7684\u4ea4\u53c9\u548c\u524d\u6cbf\u5b66\u79d1, \u5728\u5f88\u591a\u65b9\u9762\u53d6\u5f97\u4e86\u4e30\u7855\u6210\u679c.\u5c24\u5176\u662f\u8fd1\u6765\u6df1\u5ea6\u5b66\u4e60\u7684\u53d1\u5c55, \u66f4\u662f\u5c06AI\u7814\u7a76\u63a8\u5411\u4e00\u4e2a\u65b0\u7684\u9ad8\u6f6e.\u6df1\u5ea6\u5b66\u4e60\u7684\u6210\u529f\u6545\u4e8b\u6bcf\u5929\u90fd\u5728\u4e0a\u6f14, \u4f3c\u4e4e\u6df1\u5ea6\u5b66\u4e60\u662f\u65e0\u6240\u4e0d\u80fd\u7684.\u6240\u4ee5, \u4eba\u4eec\u5bf9AI\u4ea7\u751f\u6d53\u539a\u5174\u8da3, \u4ea7\u4e1a\u754c\u9996\u5148\u5e03\u5c40, \u5927\u91cf\u8d44\u672c\u4e0e\u5e76\u8d2d\u7684\u6d8c\u5165\u5e76\u52a0\u901fAI\u6280\u672f\u4e0e\u5e94\u7528\u7684\u7ed3\u5408, \u8513\u5ef6\u5347\u6e29.\u90a3\u4e48, \u629b\u5f00\u5546\u4e1a\u56e0\u7d20, AI\u5728\u6280\u672f\u4e0a\u662f\u5426\u8fd8\u5b58\u5728\u4ec0\u4e48\u95ee\u9898?\u5b83\u4eca\u540e\u7684\u53d1\u5c55\u65b9\u5411\u662f\u4ec0\u4e48? <\/p>\n \u6df1\u5ea6\u5b66\u4e60\u6280\u672f\u548c\u65e9\u5e74\u7684\u591a\u5c42\u795e\u7ecf\u7f51\u7edc\u76f8\u6bd4, \u7406\u8bba\u4e0a\u7684\u7a81\u7834\u5176\u5b9e\u4e0d\u5927.\u6df1\u5ea6\u5b66\u4e60\u73b0\u5b58\u4e00\u4e2a\u6839\u672c\u7684\u7f3a\u9677, \u5c31\u662f\u7f3a\u4e4f\u5b8c\u5584\u7684\u7406\u8bba\u652f\u6491, \u5bf9\u4e8e\u5176\u7ed3\u679c\u7f3a\u4e4f\u89e3\u91ca, \u5305\u62ec\u7f51\u7edc\u7684\u8bbe\u8ba1\u3001\u53c2\u6570\u7684\u8bbe\u7f6e\u548c\u53d6\u5f97\u7684\u6548\u679c.\u5bfb\u627e\u7406\u8bba\u4e0a\u7684\u6df1\u5165\u89e3\u91ca, \u662f\u6df1\u5ea6\u5b66\u4e60\u53d1\u5c55\u5fc5\u987b\u89e3\u51b3\u7684\u4e00\u4e2a\u91cd\u8981\u95ee\u9898. <\/p>\n \u76ee\u524d\u6df1\u5ea6\u5b66\u4e60\u4e00\u822c\u9700\u8981\u5927\u6570\u636e\u652f\u6301, \u4f46\u5e76\u4e0d\u662f\u6240\u6709\u7684\u5e94\u7528\u90fd\u5177\u5907\u5927\u6570\u636e\u6761\u4ef6\u7684.\u7ed3\u5408\u4f20\u7edf\u77e5\u8bc6\u8868\u8fbe\u548c\u6570\u636e\u9a71\u52a8\u77e5\u8bc6\u5b66\u4e60, \u53ef\u4ee5\u89e3\u51b3\u5f88\u591a\u8feb\u5207\u7684\u73b0\u5b9e\u95ee\u9898.\u8fd9\u4e5f\u662f\u4eca\u540e\u53d1\u5c55\u7684\u4e00\u4e2a\u91cd\u8981\u65b9\u5411. <\/p>\n \u8fd8\u6709, \u6df1\u5ea6\u5b66\u4e60\u5728\u8bad\u7ec3\u7f51\u7edc\u4e2d\u9700\u8981\u5927\u91cf\u6709\u6807\u8bb0\u7684\u6570\u636e\u53bb\u5b66\u4e60\u8f93\u5165\u548c\u8f93\u51fa\u7684\u6620\u5c04\u5173\u7cfb, \u8fd9\u6837\u83b7\u5f97\u7684\u6a21\u578b\u5f80\u5f80\u65e0\u6cd5\u5c06\u5176\u6cdb\u5316\u5230\u4e0e\u8bad\u7ec3\u65f6\u4e0d\u540c\u6761\u4ef6\u7684\u6570\u636e\u96c6\u4e0a.\u800c\u73b0\u5b9e\u5e94\u7528\u4e2d, \u6211\u4eec\u9047\u5230\u7684\u6570\u636e\u96c6\u5e38\u5e38\u4f1a\u5305\u542b\u5f88\u591a\u65b0\u573a\u666f, \u8bb8\u591a\u6570\u636e\u662f\u6a21\u578b\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u6ca1\u51fa\u73b0\u8fc7\u7684, \u56e0\u6b64\u5b66\u4e60\u5f97\u5230\u7684\u6a21\u578b\u53ef\u80fd\u65e0\u6cd5\u5f88\u597d\u5730\u9884\u6d4b\u7ed3\u679c.\u5c06\u5b66\u4e60\u5f97\u5230\u7684\u77e5\u8bc6\u8fc1\u79fb\u5230\u65b0\u7684\u6761\u4ef6\u548c\u73af\u5883\u7684\u80fd\u529b\u901a\u5e38\u88ab\u79f0\u4e3a\u8fc1\u79fb\u5b66\u4e60, \u8fd9\u662f\u4eca\u540e\u4e00\u4e2a\u91cd\u8981\u7684\u7814\u7a76\u65b9\u5411.\u5982\u679c\u6211\u4eec\u5c06\u8fc1\u79fb\u5b66\u4e60\u505a\u5230\u6781\u9650, \u4ec5\u4ec5\u4ece\u5c11\u6570\u51e0\u4e2a\u751a\u81f3\u96f6\u4e2a\u6837\u672c\u4e2d\u5b66\u4e60 (\u5982\u4e00\u6b21\u548c\u96f6\u6b21\u5b66\u4e60) , \u5c06\u80fd\u89e3\u51b3\u66f4\u591a\u5b9e\u9645\u95ee\u9898.\u6267\u884c\u4e00\u6b21\u548c\u96f6\u6b21\u5b66\u4e60\u7684\u6a21\u578b\u662f\u673a\u5668\u5b66\u4e60\u4e2d\u6700\u96be\u7684\u95ee\u9898\u4e4b\u4e00, \u53ef\u8fd9\u5bf9\u6211\u4eec\u4eba\u7c7b\u800c\u8a00\u5374\u4e0d\u662f\u90a3\u4e48\u56f0\u96be\u7684.\u8fd9\u662fAI\u53d1\u5c55\u4e00\u4e2a\u503c\u5f97\u6df1\u5165\u7814\u7a76\u7684\u95ee\u9898. <\/p>\n \u53e6\u5916\u6709\u4e00\u4e2a\u975e\u5e38\u503c\u5f97\u8fdb\u4e00\u6b65\u601d\u8003\u7684\u95ee\u9898:\u662f\u5426\u975e\u5f97\u8981\u91c7\u7528\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u8fdb\u884c\u5b66\u4e60?\u5982\u679c\u6240\u8003\u8651\u95ee\u9898\u7684\u6570\u636e\u548c\u76ee\u6807\u4e4b\u95f4\u7684\u51fd\u6570\u5173\u7cfb\u6bd4\u8f83\u7b80\u5355, \u90a3\u4e48\u6211\u4eec\u5b8c\u5168\u53ef\u4ee5\u7528\u6d45\u5ea6\u7684\u7f51\u7edc\u8fdb\u884c\u5efa\u6a21\u5b66\u4e60.\u4f46\u662f\u5982\u679c\u8fd9\u4e2a\u51fd\u6570\u7684\u786e\u6bd4\u8f83\u590d\u6742, \u662f\u5426\u4e00\u5b9a\u8981\u7528\u6df1\u5ea6\u7f51\u7edc\u5462?\u9488\u5bf9\u8fd9\u4e2a\u95ee\u9898, \u5357\u4eac\u5927\u5b66\u5468\u5fd7\u534e\u6559\u6388\u7b49\u63d0\u51fa\u4e00\u79cd\u57fa\u4e8e\u6811\u7684\u65b9\u6cd5, \u53eb\u201c\u6df1\u5ea6\u68ee\u6797\u201d[35]<\/a>, \u6765\u6311\u6218\u6df1\u5ea6\u5b66\u4e60.\u5728\u8bbe\u7f6e\u53ef\u7c7b\u6bd4\u7684\u60c5\u51b5\u4e0b, \u6df1\u5ea6\u68ee\u6797\u53d6\u5f97\u4e86\u548c\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u76f8\u5f53\u751a\u81f3\u66f4\u597d\u7684\u7ed3\u679c, \u800c\u4e14\u66f4\u5bb9\u6613\u8bad\u7ec3, \u5c0f\u6570\u636e\u4e5f\u80fd\u8fd0\u884c.\u66f4\u91cd\u8981\u7684\u662f\u76f8\u6bd4\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc, \u57fa\u4e8e\u6811\u7684\u65b9\u6cd5\u4e0d\u4f1a\u5b58\u5728\u90a3\u4e48\u56f0\u96be\u7684\u7406\u8bba\u5206\u6790\u95ee\u9898.\u4ed6\u4eec\u7684\u65b9\u6cd5\u4e3a\u5728\u8bb8\u591a\u4efb\u52a1\u4e2d\u4f7f\u7528\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u4e4b\u5916\u7684\u65b9\u6cd5\u6253\u5f00\u4e86\u4e00\u6247\u95e8. <\/p>\n 6 \u7ed3\u8bba<\/p>\n \u5230\u76ee\u524d\u4e3a\u6b62, \u4eba\u5de5\u667a\u80fd\u7684\u7814\u7a76\u4f9d\u7136\u8fd8\u5904\u4e8e\u521d\u7ea7\u9636\u6bb5, \u8ddd\u79bb\u6700\u7ec8\u7684\u76ee\u6807\u8fd8\u6709\u5f88\u957f\u7684\u8def\u8981\u8d70.\u6df1\u5ea6\u5b66\u4e60\u65b9\u6cd5\u53d6\u5f97\u4e86\u5de8\u5927\u7684\u8fdb\u5c55, \u4f46\u662f\u6ca1\u6709\u575a\u5b9e\u7684\u7406\u8bba\u57fa\u7840, \u65e0\u6cd5\u5b9e\u73b0\u5bf9\u7cfb\u7edf\u548c\u6027\u80fd\u7684\u900f\u5f7b\u7406\u89e3\u548c\u9884\u6d4b.\u8fd8\u6709\u5f88\u591a\u7684\u95ee\u9898\u6446\u5728\u6211\u4eec\u9762\u524d, \u5982\u8fc1\u79fb\u5b66\u4e60\u3001\u5c0f\u6837\u672c\u5b66\u4e60\u3001\u589e\u5f3a\u5b66\u4e60\u7b49, \u8fd9\u4e9b\u90fd\u662f\u4eba\u5de5\u667a\u80fd\u7814\u7a76\u6025\u9700\u89e3\u51b3\u7684\u95ee\u9898.\u867d\u7136\u4eba\u5de5\u667a\u80fd\u5728\u5f88\u591a\u65b9\u9762\u5df2\u7ecf\u8d76\u8d85\u4eba\u7c7b, \u4f46\u662f\u672c\u8d28\u4e0a\u79bb\u771f\u6b63\u7684\u667a\u80fd\u8fd8\u662f\u6709\u5f88\u5927\u7684\u8ddd\u79bb, \u8fd9\u4e5f\u662f\u6211\u4eec\u8fdb\u4e00\u6b65\u671f\u5f85\u548c\u52aa\u529b\u7684\u65b9\u5411. <\/p>\n \u53c2\u8003\u6587\u732e<\/p>\n [1]\u674e\u857e, \u738b\u5c0f\u6377.\u673a\u5668\u667a\u80fd[M].\u5317\u4eac:\u6e05\u534e\u5927\u5b66\u51fa\u7248\u793e, 2016LI Lei, WANG Xiaojie.Machine intelligence[M].Beijing:Tsinghua University Press, 2016<\/a> <\/p>\n [2]\u987e\u9669\u5cf0.\u4eba\u5de5\u667a\u80fd\u7684\u5386\u53f2\u56de\u987e\u548c\u53d1\u5c55\u73b0\u72b6[J].\u81ea\u7136\u6742\u5fd7, 2016, 38 (3) :157-166GU Xianfeng.Historical review and current development of artificial intelligence[J].Chinese Journal of Nature, 2016, 38 (3) :157-166<\/a> <\/p>\n [3]Hinton G, Osindero S, Teh Y W.A fast learning algorithm for deep belief nets[J].Neural Computation, 2006, 18 (7) :1527-1554<\/a> <\/p>\n [4]Hinton G, Salakhutdinov R R.Reducing the dimensionality of data with neural networks[J].Science, 2006, 313 (5786) :504-507<\/a> <\/p>\n [5]\u949f\u4e49\u4fe1.\u9ad8\u7b49\u4eba\u5de5\u667a\u80fd\u539f\u7406:\u89c2\u5ff5\u3001\u65b9\u6cd5\u3001\u6a21\u578b\u3001\u7406\u8bba[M].\u5317\u4eac:\u79d1\u5b66\u51fa\u7248\u793e, 2014ZHONG Yixin.Principles of advanced artificial intelligence:Concepts, methods, models and theories[M].Beijing:Science Press, 2014<\/a> <\/p>\n [6]Mc Culloch W S, Pitts W.A logical calculus of the ideas immanent in nervous activity[J].Bulletin of Mathematical Biology, 1990, 52 (1\/2) :99-115<\/a> <\/p>\n [7]Rosenblatt F.The perceptron:A probabilistic model for information storage and organization in the brain[J].Psychological Review, 1958, 65 (6) :386-408<\/a> <\/p>\n [8]Haykin S.Neural networks:A comprehensive foundation[M].Upper Saddle River, NJ:Prentice Hall, 2004<\/a> <\/p>\n [9]Lindsay R K, Buchanan B G, Feigenbaum E A, et al.DENDRAL:A case study of the first expert system for scientific hypothesis formation[J].Artificial Intelligence, 1993, 61 (2) :209-261<\/a> <\/p>\n [10]Shortliffe E H.Mycin:A rule-based computer program for advising physicians regarding antimicrobial therapy selection[D].Stanford:Stanford University, 1975<\/a> <\/p>\n [11]Brooks R A.Intelligence without representation[J].Artificial Intelligence, 1991, 47 (1\/2\/3) :139-159<\/a> <\/p>\n [12]Rumelhart D E, Hinton G, Williams R J.Learning internal representations by error propagation[J].Readings in Cognitive Science, 1988, 1:399-421<\/a> <\/p>\n [13]Rumelhart D E, Hinton G, Williams R J.Learning representations by back-propagating errors[J].Nature, 1986, 323 (6088) :533-536<\/a> <\/p>\n [14]Hornik K.Approximation capabilities of multilayer feedforward networks[J].Neural Networks, 1991, 4 (2) :251-257<\/a> <\/p>\n [15]Csaji B C.Approximation with artificial neural networks[D].Budapest, Hungary:Faculty of Sciences, Etvs Lornd University, 2001<\/a> <\/p>\n [16]Le Cun Y, Boser B, Denker J S, et al.Backpropagation applied to handwritten zip code recognition[J].Neural Computation, 1989, 1 (4) :541-551<\/a> <\/p>\n [17]Le Cun Y, Bottou L, Bengio Y, et al.Gradient-based learning applied to document recognition[J].Proceedings of IEEE, 1998, 86 (11) :2278-2324<\/a> <\/p>\n [18]Hochreiter S.The vanishing gradient problem during learning recurrent neural nets and problem solutions[J].International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems, 1998, 6 (2) :107-116<\/a> <\/p>\n [19]Vapnik V.The nature of statistical learning theory[M].New York:Springer, 1999<\/a> <\/p>\n [20]Glorot X, Bordes A, Bengio Y.Deep sparse rectifier networks[J].Journal of Machine Learning Research, 2011 (15) :315-323<\/a> <\/p>\n [21]Hinton G, Srivastava N, Krizhevsky A, et al.Improving neural networks by preventing co-adaptation of feature detectors[J].Computer Science, 2012, 3 (4) :212-223<\/a> <\/p>\n [22]Raina R, Madhavan A, Ng A Y.Large-scale deep unsupervised learning using graphics processors[C]\u2225Proceedings of the 26th Annual International Conference on Machine Learning, 2009:873-880<\/a> <\/p>\n [23]Git Hub, Inc.Deep-learning-papers-reading-roadmap[EB\/OL].[2017-04-28].https:\u2225github.com\/songrotek\/Deep-Learning-Papers-Reading-Roadmap<\/a> <\/p>\n [24]Sharma J, Angelucci A, Sur M.Induction of visual orientation modules in auditory cortex[J].Nature, 2000, 404 (6780) :841-847<\/a> <\/p>\n [25]Vuillerme N, Cuisinier R.Sensory supplementation through tongue electrotactile stimulation to preserve head stabilization in space in the absence of vision[J].Investigative Ophthalmology&Visual Science, 2009, 50 (1) :476-481<\/a> <\/p>\n [26]Le Cun Y, Bengio Y, Hinton G.Deep learning[J].Nature, 2015, 521 (7553) :436-444<\/a> <\/p>\n [27]Deng J, Dong W, Socher R, et al.Image Net:A large-scale hierarchical image database[C]\u2225IEEE Conference on Computer Vision and Pattern Recognition, 2009:248-255<\/a> <\/p>\n [28]Hinton G, Deng L, Yu D, et al.Deep Neural Networks for acoustic modeling in speech recognition:The shared views of four research groups[J].IEEE Signal Processing Magazine, 2012, 29 (6) :82-97<\/a> <\/p>\n [29]Graves A, Mohamed A, Hinton G.Speech recognition with deep recurrent neural networks[C]\u2225IEEE International Conference on Acoustics, Speech and Signal Processing, 2013, DOI:10.1109\/ICASSP.2013.6638947<\/a> <\/p>\n [30]Esteva A, Kuprel B, Novoa R A, et al.Dermatologist-level classification of skin cancer with deep neural networks[J].Nature, 2017, 542 (7639) :115<\/a> <\/p>\n [31]Hutson M.Self-taught artificial intelligence beats doctors at predicting heart attacks[EB\/OL].Science, 2017-04-14[2017-04-28].http:\u2225www.sciencemag.org\/news\/2017\/04\/self-taught-artificial-intelligence-beats-doctors-predicting-heart-attacks<\/a> <\/p>\n [32]Silver D, Huang A, Maddison C J, et al.Mastering the game of Go with deep neural networks and tree search[J].Nature, 2016, 529 (7587) :484-489<\/a> <\/p>\n [33]Mnih V, Kavukcuoglu K, Silver D, et al.Human-level control through deep reinforcement learning[J].Nature, 2015, 518 (7540) :529-533<\/a> <\/p>\n [34]Gatys L A, Ecker A S, Bethge M.A neural algorithm of artistic style[J].ar Xiv, 2015, ar Xiv:1508.06576 v1<\/a> <\/p>\n [35]Zhou Z H, Feng J.Deep forest:Towards an alternative to deep neural networks[J].ar Xiv, 2017, ar Xiv:1702.08835v1<\/a><\/p>\n","protected":false},"excerpt":{"rendered":" \u4fde\u795d\u826f \u6458 \u8981\uff1a\u8fd1\u5e74\u6765, \u4ee5\u6df1\u5ea6\u5b66\u4e60\u4e3a\u6838\u5fc3\u7684\u4eba\u5de5\u667a\u80fd\u6280\u672f, \u53d6\u5f97\u4e86\u4e00\u7cfb\u5217\u91cd\u5927\u7a81\u7834 … \u7ee7\u7eed\u9605\u8bfb