{"id":78010,"date":"2021-03-05T15:01:00","date_gmt":"2021-03-05T07:01:00","guid":{"rendered":"https:\/\/www.wingwill.com.tw\/?p=18944"},"modified":"2025-11-27T09:38:00","modified_gmt":"2025-11-27T01:38:00","slug":"blog-a61-transfer-learning","status":"publish","type":"products","link":"https:\/\/www.wingwill.com.tw\/ja\/products\/blog-a61-transfer-learning\/","title":{"rendered":"Transfer learning \u521d\u9ad4\u9a57"},"content":{"rendered":"

\u4e0d\u66c9\u5f97\u4f60\u6709\u6c92\u6709\u60f3\u904e\u6bd4\u8d77\u5176\u4ed6\u98df\u7269\u93c8\u9802\u7aef\u7684\u5176\u4ed6\u751f\u7269\uff0c\u70ba\u4f55\u53ea\u6709\u4eba\u985e\u53ef\u4ee5\u767c\u5c55\u9ad8\u5ea6\u6587\u660e\uff0c\u751a\u81f3\u53ef\u4ee5\u822a\u5411\u5b87\u5b99\u661f\u8fb0\uff1f\u7b54\u6848\u5c31\u662f\u5178\u7bc4\u79fb\u8f49\uff01\u672c\u6587\u5c07\u6703\u4ecb\u7d39\u5728AI\u7684\u4e16\u754c\u4e86\u5982\u4f55\u61c9\u7528\u9019\u500b\u6280\u8853\uff0c\u7528\u6700\u77ed\u7684\u6642\u9593\u8207\u6700\u5c0f\u7684\u8cc7\u6599\u96c6\uff0c\u8a13\u7df4\u51fa\u4e0d\u932f\u7684AI\u6a21\u578b\u2026<\/p>\n\n\n\n

\"pretrained-model\"<\/figure>\n\n\n\n

\u524d\u8a00\uff1a
<\/strong>\u6709\u670b\u53cb\u554f\u6211\u300c\u53f0\u7063\u7684\u4e2d\u5c0f\u4f01\u696d\u4f54\u4e86 95%\uff0c\u5927\u5bb6\u90fd\u990a\u4e0d\u8d77\u8cc7\u6599\u79d1\u5b78\u5bb6\uff0c\u96a8\u8457 AI \u7684\u6280\u8853\u6108\u4f86\u6108\u6210\u719f\uff0c\u96e3\u9053\u6c92\u6709\u66f4\u5bb9\u6613\u8b93AI\u666e\u53ca\u7684\u65b9\u6cd5\u55ce\uff1f\u300d\u5c31\u50cf\u4f60\u60f3\u5207\u5165\u6c7d\u8eca\u5e02\u5834\uff0c\u5728\u6c7d\u8eca\u5de5\u696d\u9019\u9ebc\u767c\u9054\u7684\u73fe\u4eca\uff0c\u61c9\u8a72\u4e0d\u9700\u8981\u5f9e\u505a\u8f2a\u5b50\u958b\u59cb\u5427\uff1f\u4e0d\u7136\u7d14\u96fb\u52d5\u8eca\u7684\u7279\u65af\u62c9\uff0c\u662f\u600e\u9ebc\u6a6b\u7a7a\u51fa\u4e16\u7684\uff1f
\u5176\u5be6\u9019\u500b\u554f\u984c\uff0c\u4e0d\u597d\u56de\u7b54\uff0c\u4f46\u5728\u6211\u5fc3\u4e2d\u6709\u4e09\u7a2e\u8173\u672c\u7684\u7b54\u6848\uff0c\u65b9\u6848\u4e00\u53eb Auto ML\uff0c\u8981\u670d\u52d9\u7684\u5c0d\u8c61\u662f\u6709\u9322\u4eba\u5bb6\u7684\u516c\u5b50\u723a\u5011\uff1b\u65b9\u6848\u4e8c\u53ebAI gallery\uff0c\u662f\u6709GPU\u7684\u958b\u767c\u5718\u968a\uff1b\u65b9\u6848\u4e09\u53ebTransfer Learning\uff0c\u662f\u6709RD\u5718\u968a\u3002
\u5728 Azure Auto ML<\/a>\u7684\u89e3\u6c7a\u65b9\u6848\u88e1\uff0c\u53ea\u8981\u4f60\u6709\u660e\u78ba\u7684\u984c\u76ee\u8207\u8cc7\u6599\u96c6\uff0c\u5b8c\u5168\u4e0d\u9700\u8981\u7ba1 AI\u6d41\u7a0b\u4e2d\u7684\u8cc7\u6599\u6574\u7406\u3001\u7279\u5fb5\u5de5\u7a0b\u3001\u6f14\u7b97\u6cd5\u9078\u64c7\u3001\u8d85\u53c3\u6578\u8abf\u6574\u2026\u7b49\u7d30\u7bc0(\u5982\u679c\u5c0d\u5b8c\u6574\u7684 AI\u6d41\u7a0b\u4e0d\u719f\u6089\u53ef\u4ee5\u53c3\u8003\u9019\u4e00\u7bc7<\/a>)\uff0c\u5c31\u80fd\u5728 Portal \u4e0a\u8f15\u9b06\u7684\u4e00\u9375\u751f\u6210\u6a21\u578b\u3002\u56e0\u70ba\uff0c\u96f2\u7aef\u5e73\u53f0\u6703\u5728\u80cc\u5f8c\u4ee5\u66b4\u529b\u6cd5\uff0c\u9ad8\u4f75\u767c\u7684\u65b9\u5f0f\u5e6b\u4f60\u5690\u8a66\u5404\u7a2e\u6392\u5217\u7d44\u5408\uff0c\u7136\u5f8c\u4f60\u5c31\u53ef\u4ee5\u770b\u5831\u8868\uff0c\u6c7a\u5b9a\u4f60\u8981\u7528\u54ea\u500b\u6a21\u578b\uff1f
\u5728 Azure AI Gallery<\/a>\u89e3\u6c7a\u65b9\u6848\u88e1\uff0c\u9664\u4e86\u4f60\u9700\u8981\u6709\u660e\u78ba\u7684\u984c\u76ee\u8207\u8cc7\u6599\u96c6\u4e4b\u5916\uff0c\u4f60\u9700\u8981\u5c0d\u5b8c\u6574\u7684 AI\u6d41\u7a0b\u6709\u4e00\u4e9b\u6982\u5ff5\u3002\u56e0\u70ba\u7576\u4f60\u7528\u5206\u985e\u7d22\u5f15\u7684\u700f\u89bd\u6216\u662f\u95dc\u9375\u5b57\u641c\u5c0b\u5230\u9069\u5408\u4f60\u7684 AI template\u4e4b\u5f8c\uff0c\u4f60\u9084\u9700\u8981\u6709 Azure \u8a02\u95b1\uff0c\u624d\u80fd\u958b\u555f AI \u7bc4\u672c\uff0c\u628a\u8cc7\u6599\u4f86\u6e90\u7f6e\u63db\u6210\u4f60\u7684 Data set\uff0c\u5c0d\u6d41\u7a0b\u5167\u7684\u7d30\u7bc0\u4ee5\u62d6\u62c9\u9ede\u9078\u7684\u65b9\u6848\uff0c\u624d\u80fd\u50cf\u500b\u82b1\u5320\u4e00\u6a23\u4fee\u679d\u6f64\u98fe\uff0c\u5b8c\u6210\u8abf\u6574\u8d85\u53c3\u6578\u6216\u662f\u640d\u5931\u51fd\u5f0f\u9078\u64c7\u7684\u76f8\u95dc\u52d5\u4f5c\u3002
\u5982\u679c\u4e0a\u8ff0\u9084\u4e0d\u80fd\u6eff\u8db3\u4f60\uff0c\u5728\u5fae\u8edf\u5ee3\u5927\u7684AI \u751f\u614b\u7cfb\u4e2d\u9084\u6709 ML.Net<\/a>, DataBricks<\/a> …\u591a\u7a2e\u89e3\u6c7a\u65b9\u6848\u4e2d\uff0c\u9664\u4e86\u4f60\u9700\u8981\u6709\u660e\u78ba\u7684\u984c\u76ee\u8207\u8cc7\u6599\u96c6\u4e4b\u5916\uff0c\u4f60\u9084\u9700\u8981\u5c0d\u5b8c\u6574\u7684 AI\u6d41\u7a0b\u6709\u719f\u7df4\u7684\u898f\u5283\u8207\u64cd\u4f5c\u80fd\u529b\uff0c\u4e26\u77e5\u9053 Transfer learning \u7684\u539f\u7406\u8207\u904b\u7528\u7684\u65b9\u6cd5\uff0c\u4f8b\u5982\u4f60\u8981\u4e0b\u8f09\u7531\u5b78\u754c\u77e5\u540d\u8cc7\u6599\u96c6(Coco, ImageNet…)\u6240\u8a13\u7df4\u51fa\u4f86\u7684\u6a21\u578b\u96c6(\u591a\u5927\u898f\u6a21\u7684\u6a21\u578b\uff0c\u4f8b\u5982 ResNet50\/101\/152…)\uff0c\u662f\u8981\u9077\u79fb\u5168\u90e8 Layer\u7684\u53c3\u6578\uff0c\u9084\u662f\u90e8\u4efd Layer\uff1f  <\/p>\n\n\n\n

\"Automated<\/figure>\n\n\n\n

\u4f55\u8b02\u9077\u79fb\u5f0f\u5b78\u7fd2\uff1f
<\/strong>\u4e0d\u66c9\u5f97\u4f60\u6709\u6c92\u6709\u9019\u6a23\u7684\u7d93\u9a57\uff0c\u8981\u6559\u4e00\u500b\u5973\u751f\u5b78\u6703\u9a0e\u6a5f\u8eca\uff0c\u53ef\u4ee5\u5148\u6559\u5979\u9a0e\u8173\u8e0f\u8eca\uff1f\u5728\u5b78\u8853\u754c\uff0c\u662f\u5148\u7531\u5fc3\u7406\u5b78\u5bb6\u6295\u5165\u9019\u6a23\u7684\u7814\u7a76\uff0c\u63a5\u8457\u8cc7\u8a0a\u79d1\u5b78\u5bb6\uff0c\u958b\u555f\u4e86\u4f9d\u64da\u9077\u79fb\u7684\u60c5\u5883(Scenario)\u3001\u65b9\u6cd5(Method)\u3001\u7279\u5fb5\u7a7a\u9593(Feature space)\u7684\u5c08\u9580\u7814\u7a76\uff0c
\u82e5\u4f60\u8981\u6211\u4e0b\u4e00\u500b\u5b9a\u7fa9\uff0c\u5c31\u50cf\u4eba\u985e\u53ef\u4ee5\u8209\u4e00\u53cd\u4e09\uff0c\u53ea\u8981\u8b93AI \u770b\u904e\u8db3\u5920\u7684 data\uff0c\u5b78\u6703\u4f4e\u7dad\/\u4e2d\u7dad\/\u9ad8\u7dad\u7684 Pattern\uff0c\u5b83\u9664\u4e86\u53ef\u4ee5\u8655\u7406\u597d\u7576\u524d\u7684\u984c\u76ee\uff0c\u751a\u81f3\u4e5f\u53ef\u4ee5\u900f\u904e\u5178\u7bc4\u79fb\u8f49\u7684\u65b9\u5f0f\uff0c\u5c07\u53c3\u6578\u79fb\u82b1\u63a5\u6728\u81f3\u53e6\u4e00\u500b AI\u6a21\u578b\uff0c\u8655\u7406\u4e0d\u540c\u7684\u984c\u76ee\u3002
\u5728\u9077\u79fb\u5f0f\u5b78\u7fd2\u7684\u6846\u67b6\u4e2d\uff0c\u7531\u4e0a\u800c\u4e0b\uff0c\u5206\u5225\u662f\uff1aDomain, Task,  Label, Class\uff0c\u53ef\u4ee5\u662f\u55ae\u7d14\u7684\u554f\u984c\uff1b\u4e5f\u53ef\u4ee5\u662f\u8907\u96dc\u7684\u554f\u984c Multi-task (\u4f8b\u5982\u5c07\u8c93\u72d7\u5206\u985e\u5668\u8207\u8001\u864e\u5927\u8c61\u5206\u985e\u5668\u4e00\u8d77\u8a13\u7df4)\/ Multi-label (\u4f8b\u5982\u6cb9\u96fb\u6df7\u5408\u8eca\uff0c\u540c\u6642\u6a19\u8a18\u88ab\u6a19\u8a18\u6c7d\u6cb9\u8eca & \u96fb\u52d5\u8eca\u4e8c\u500b Label)\/ Multi-class (\u4f8b\u5982\u6cb9\u96fb\u6df7\u5408\u8eca\uff0c\u4ed6\u540c\u6642\u6703\u9810\u6e2c\u70ba\u6c7d\u6cb9\u8eca & \u96fb\u52d5\u8eca\u4e8c\u500b\u5206\u985e)\u7684\u5206\u985e\u554f\u984c\uff0c\u900f\u904e shared representation(\u5bb6\u8c93\u8207\u8001\u864e\u90fd\u662f\u8c93\u79d1\uff0c\u8cc7\u8a0a\u53ef\u4ee5\u5171\u4eab) \u4e00\u8d77\u8a13\u7df4\uff0c\u4f86\u9054\u5230\u7701\u6642\u7701\u529b\u7684\u9ad8\u6548\u6210\u679c\u3002<\/p>\n\n\n\n

\u53f0\u5927\u7684\u674e\u5b8f\u6bc5\u6559\u6388\u5c31\u66fe\u7d93\u8209\u904e\u9019\u6a23\u7684\u4f8b\u5b50\uff1a\u6709\u4e00\u500b\u9ad8\u624b\u5b78\u751f\uff0c\u61c9\u7528\u4e86\u5b78\u8853\u754c\u77e5\u540d\u5f71\u50cf\u8cc7\u6599\u96c6\uff0c\u505a\u51fa\u4e86\u8c93\u72d7\u5206\u985e\u5668\uff0c\u5b8c\u6210\u4e86\u671f\u4e2d\u4f5c\u696d\u4e26\u4e14\u5f97\u5230\u9ad8\u5206\u3002\u63a5\u4e0b\u4f86\u8001\u5e2b\u53c8\u51fa\u4e86\u671f\u672b\u7684\u4f5c\u696d\uff0c\u4ed6\u904b\u6c23\u5f88\u597d\uff01\u88ab\u5206\u5230\u76f8\u4f3c\u7684\u8c93\u72d7\u5206\u985e\u984c\u76ee\uff0c\u53ea\u4e0d\u904e\u9019\u6b21\u7684\u8cc7\u6599\u96c6\u88ab\u63db\u6210\u5361\u901a\u7684\uff0c\u65bc\u662f\u4ed6\u900f\u904e Pre-train & Fine tuning \u624b\u6cd5\uff0c\u5c07\u524d\u4e00\u500b AI\u6a21\u578b\u7684\u53c3\u6578\u79fb\u8f49\u7d66\u65b0\u7684 AI\u6a21\u578b\uff0c\u8f15\u6613\u5730\u5b8c\u6210\u9ad8\u98db\u72d7(Goofy)\u8207\u62db\u8ca1\u8c93\u7684\u8fa8\u8b58\uff0c\u9019\u500b\u53eb Different domain, same task\u3002\u9ad8\u624b\u559c\u6b61\u5e6b\u52a9\u7570\u6027\uff0c\u65bc\u662f\u4ed6\u6539\u4e86\u4e00\u4e0b\u7a0b\u5f0f\uff0c\u4e5f\u6210\u529f\u5730\u5b8c\u6210\u5973\u540c\u5b78\u88ab\u5206\u5230\u7684\u975e\u6d32\u5927\u8349\u539f\u4e0a\u7684\u5927\u8c61\u8207\u8001\u864e\u7684\u5206\u985e\u5668\uff0c\u9019\u500b\u53eb Same domain, different task\u3002
\u7e3d\u4e4b\uff0c\u674e\u6559\u6388\u7e3d\u7d50\u4e86\u6700\u8fd1\u570b\u5167\u5916\u7684\u5091\u51fa\u5b78\u8005\u7684\u7814\u7a76\u6210\u679c\uff0c\u4f9d\u64da Source \u7684 Data \u6709\u6a19\u8a18\/\u7121\u6a19\u8a18\uff0c\u4f7f\u7528\u8207Target \u7684 data \u6709\u6a19\u8a18\/\u7121\u6a19\u8a18\uff0c\u6a6b\u8de8\u4e86\u76e3\u7763\u5f0f\/\u975e\u76e3\u7763\u5f0f\u5b78\u7fd2\u5728\u56db\u500b\u8c61\u9650\u4e2d\u5171\u6709\u6578\u7a2e\u56e0\u61c9\u7684\u65b9\u6cd5\uff1a<\/p>\n\n\n\n

<\/td>Source data labelled<\/td>Source data unlabeled<\/td><\/tr>
Target data labelled<\/td>1.Fine tuning
2. Multitask learnin<\/td>		5.Self-taught learning<\/td><\/tr>
Targe data unlabeled<\/td>3.Domain-adversarial training
4. Zero-shot learning<\/td>		6.Self-taught clustering<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
  1. Fine tuning\uff1a\u4ee5CNN\u61c9\u7528\u5728\u5f71\u50cf\u8fa8\u8b58\u70ba\u4f8b\uff0c\u6307\u7684\u662f\u51cd\u7d50\u524d\u5e7e\u5c64Layer \u53c3\u6578\uff0c\u53ea\u8b93\u6b64\u6b21\u65b0\u7684 Data \u53c3\u8207\u5f8c\u5e7e\u5c64Layer \u7684\u8a13\u7df4\u3002\u4f60\u53ef\u80fd\u6703\u554f\u8aaa\u70ba\u4f55\u53ea\u56fa\u5b9a\u524d\u9762\u5e7e\u5c64\uff1f\u662f\u56e0\u70ba CNN\u7684\u5b78\u7fd2\u662f\u6709\u5c64\u6b21\u7684\uff0c\u662f\u7531\u9ede\u3001\u7dda\u3001\u9762\u6162\u6162\u5b78\u8d77\u4f86\u7684\uff0c\u4e0d\u7ba1\u662f\u8fa8\u8b58\u52d5\u7269\u3001\u690d\u7269\u3001\u7926\u7269\uff0c\u5e95\u5c64\u90fd\u662f\u76f8\u540c\u7684\uff0c\u8981\u5230\u9ad8\u968e\u7406\u89e3\u6642\uff0c\u8fa8\u8b58\u52d5\u7269\u624d\u80fd\u7528\u773c\u775b\u9019\u7a2e\u4e2d\u968e\u7269\u4ef6\u4f86\u5340\u5206\u4e0d\u540c\u7684\u52d5\u7269\u3002\u6216\u8005\uff0c\u4f60\u53ef\u4ee5\u7406\u89e3\u6210\uff0c\u6211\u5011\u9700\u8981\u66fe\u7d93\u53bb\u8e72\u904e\u5c11\u6797\u5bfa\u7684\u82e6\u884c\u50e7\u5c0f\u5b69\uff0c\u63a5\u8457\u518d\u6559\u4ed6\u6700\u6642\u5c1a\u7684\u9ad8\u96e3\u5ea6\u52d5\u4f5c\uff0c\u7136\u5f8c\u4f60\u5c31\u53ef\u4ee5\u958b\u8ce3\u592a\u967d\u99ac\u6232\u5718\u7684\u7968\uff0c\u958b\u5fc3\u7684\u6578\u9214\u7968\u4e86\u3002<\/li>
  2. Multitask learning\uff1a\u9867\u540d\u601d\u7fa9\uff0cMTL\u5c31\u662f\u628a\u591a\u500b\u76f8\u95dc\u7684\u4efb\u52d9\uff0c\u4e00\u8d77\u505a\u8a13\u7df4\u3002\u5728\u76ee\u524d\u5927\u591a\u7684 AI\u61c9\u7528\u4e2d\uff0c\u591a\u534a\u90fd\u662f\u63a1\u7528\u4e86 Single task \u4f86\u8655\u7406\u554f\u984c\u3002\u4e5f\u5c31\u662f\u5c07\u539f\u672c\u8907\u96dc\u7684\u554f\u984c\u62c6\u89e3\u70ba\u7c21\u55ae\u800c\u76f8\u4e92\u7368\u7acb\u7684\u5b50\u554f\u984c\uff0c\u7136\u5f8c\u518d\u5408\u4f75\u5176\u7d50\u679c\uff0c\u8a66\u5716\u89e3\u6c7a\u8907\u96dc\u7684\u554f\u984c\u3002\u770b\u4f3c\u5408\u7406\uff0c\u537b\u5b58\u5728\u4f60\u7121\u6cd5\u4fdd\u8b49\u7368\u7acb\u5b50\u554f\u984c\u5f7c\u6b64\u4e0d\u5b58\u5728\u4e0d\u53ef\u5207\u5272\u7684\u672a\u7206\u5f48\uff0c\u70ba\u6b64\u9084\u6709\u4e00\u652f\u900f\u904e\u5171\u4eab\u4e2d\u9593\u8cc7\u8a0a\uff0c\u5c08\u9580\u5728\u8655\u7406\u8907\u96dc\u554f\u984c\u7684\u5206\u652f\u3002\u525b\u624d\u6709\u63d0\u5230\uff0c\u53c8\u53ef\u4ee5\u7d30\u5206\u70ba\u591a\u4efb\u52d9\u3001\u4e00\u500b\u7269\u4ef6\u88ab\u6a19\u8a18\u70ba\u591a\u500b\u985e\u5225\u3001\u4e00\u500b\u7269\u4ef6\u88ab\u9810\u6e2c\u70ba\u591a\u500b\u985e\u5225\u2026
    \u5728\u6a21\u578b\u7684\u8a2d\u8a08\u4e0a\uff0c\u5247\u6703\u4f9d\u64da MTL\u7684 similarity\u76f8\u4f3c\u6027, relationship\u95dc\u806f\u6027, hierrarchy\u968e\u5c64\u95dc\u4fc2, benefit\u6548\u76ca\uff0c\u5206\u5225\u53bb\u9078\u64c7 Fully-Adaptive Feature Sharing\u3001Cross-stitch Networks\u3001Low supervision\u3001A Joint Many-task Model\u3001Weighting losses with uncertainty\u3001Tensor factorisation for MTL\u3001Sluice Networks\u2026\u4e0d\u540c\u7684\u67b6\u69cb<\/li>
  3. Domain-adversarial training\uff1a\u4ee5\u6a21\u578b\u4f86\u770b\uff0c\u5b83\u5c6c\u65bc Domain-Adversarial Training of Neural Networks (DANN)\u985e\u578b\uff0c\u662f\u7528\u5728\u89e3\u6c7a\u76e3\u7763\u5f0f\u5b78\u7fd2\u9047\u5230\u6c92\u6709 Label \u7684 Data \u7684\u554f\u984c\u3002\u5b83\u7684\u539f\u7406\u662f\u61c9\u7528\u4e86\u751f\u6210\u5c0d\u6297\u7db2\u8def GAN(\u5e38\u88ab\u7528\u5728\u77e5\u540d\u96fb\u5f71\u3001\u4e09\u7d1a\u7247\u3001Kuso\u65b0\u805e\u4e0a\u8b8a\u81c9\u7684\u6f14\u7b97\u6cd5)\u7684\u624b\u6cd5\uff0c\u4f86\u89e3\u6c7a source domain \u6709 label \u4f46\u662f target domain \u6c92\u6709 label \u7684\u72c0\u6cc1\u3002\u4f46\u662f\uff0c\u5728\u5be6\u52d9\u4e2d GAN\u7684\u672c\u8cea\u4e2d\uff0c\u9664\u4e86\u9700\u8981\u8017\u8cbb\u8f03\u591a\u7684 GPU\u6210\u672c\uff0c\u5728\u4e2d\u9593\u8abf\u6821\u8d85\u53c3\u6578\u6642\u4e5f\u4e0d\u662f\u90a3\u9ebc\u76f4\u89ba\u4e14\u5bb9\u6613\u7684\uff0c\u6240\u4ee5\uff0c\u6703\u88ab\u6211\u6b78\u5728\u6709\u9322\u4eba\u7684\u65b9\u6848\u3002 <\/li>
  4. Zero-shot learning\uff1a\u76ee\u524d ZSL\u5df2\u6210\u529f\u7684\u61c9\u7528\u5728\u5f71\u50cf\u5206\u985e\u3001\u7269\u4ef6\u5075\u6e2c\u3001\u5f71\u50cf\u751f\u6210\u3001\u8a9e\u7fa9\u5206\u985e\u3001NLP\u2026\u7b49\u7bc4\u7587\u4e2d\u3002\u5728\u4e00\u958b\u59cb\uff0c\u6211\u5c31\u63d0\u5230\u904e\uff0c\u4eba\u985e\u80fd\u904b\u7528\u63a8\u7406\u4f86\u8209\u4e00\u53cd\u4e09\uff0c\u5728\u4e0d\u9700\u8981\u770b\u904e\u6240\u6709\u751f\u7269\u7684\u60c5\u6cc1\u4e0b\uff0c\u80fd\u5920\u6b63\u78ba\u5730\u5340\u5206\u51fa\u5916\u661f\u4eba\u8207\u5730\u7403\u751f\u7269\u7684\u4e0d\u540c\u3002\u4eba\u985e\uff0c\u751a\u81f3\u4e5f\u80fd\u5c07\u4e00\u4e9b\u5206\u985e\u7d50\u679c\u505a\u7d2f\u52a0\uff0c\u4f8b\u5982\u80fd\u5206\u8fa8\u751f\u7269\u8207\u975e\u751f\u7269\uff1b\u80fd\u5206\u8fa8\u6c34\u4e2d\u3001\u9678\u5730\u3001\u5929\u4e0a\u751f\u7269\uff0c\u80fd\u5206\u8fa8\u56db\u8db3\u3001\u76f4\u7acb\u3001\u722c\u884c\u7684\u9678\u5730\u751f\u7269\uff1b\u5c31\u80fd\u5340\u5206\u51fa\u4eba\u985e\u8207\u5176\u4ed6\u751f\u7269\u7684\u4e0d\u540c\u3002\u540c\u7406\uff0cZSL\u53ef\u4ee5\u76f4\u63a5\u5c0d\u4e0a\u8ff0\u7684\u4e09\u500b\u5206\u985e\u5668\u900f\u904e Knowledge transfer \u7684\u624b\u6cd5\uff0c\u5728\u4e0d\u9700\u8981\u6e96\u5099\u4eba\u985e\u7167\u7247\uff0c\u5c31\u80fd\u8a13\u7df4\u51fa\u4e00\u500b\u4eba\u985e\u5206\u985e\u5668\u3002\u53ea\u8981\u4f60\u6c92\u6709\u9047\u4e0a domain shift problem, bubness problem, semantic gap \u4e09\u500b\u7f69\u9580\u3002<\/li>
  5. Self-taught learning\uff1a\u53c8\u7a31\u70ba Soft-taught learning\uff0c\u4e3b\u8981\u662f\u900f\u904e\u975e\u76e3\u7763\u5f0f\u5b78\u7fd2\u4f86\u5b78\u6703\u7279\u5fb5\u503c\u7684\u62bd\u53d6\uff0c\u7136\u5f8c\u518d\u5c07\u6210\u679c\u7e7c\u7e8c\u7528\u5728\u76e3\u7763\u5f0f\u5b78\u7fd2\uff0c\u4f86\u89e3\u6c7a\u624b\u908a\u662f\u6c92\u6709 Label \u904e\u7684 Data\u554f\u984c\u3002\u4f8b\u5982\u5728Andrew Ng \u5927\u5e2b\u7684\u7d93\u5178\u793a\u7bc4<\/a>\u4e2d\uff0c\u7d93\u904e\u6e2c\u8a66\u5728 Minst \u624b\u5beb\u6578\u5b57\u8fa8\u8b58\u554f\u984c\u4e0a\uff0c\u539f\u51486\u5206\u9418\u53ef\u4ee5\u9054\u5230 95%\u5de6\u53f3\u7684\u6c34\u6e96\uff0c\u518d\u52a0\u5165\u4e86 Self-taught learning \u624b\u6cd5\u91cd\u65b0\u8a13\u7df4\uff0c\u7d93\u904e\u4e8630\u5206\u9418\u7684\u6642\u9593\uff0c\u53ef\u4ee5\u518d\u62c9\u9ad8\u81f3 98%\u3002\u9019\u500b\u7bc4\u4f8b\u662f\u5148\u5c070~9 \u5207\u5206\u6210 0~4\u662f LabledSet\uff1b5~9\u662f UnlabledSet\u6545\u610f\u62ff\u6389\u3002\u7136\u800c\u5728\u5be6\u52d9\u4e0a\uff0c\u7576\u4f60\u7684 Labled \u8cc7\u6599\u5f88\u5c11\uff0c\u5c31\u53ef\u4ee5\u53bb\u8cb7\uff0c\u5982\u679c\u5acc\u8cb4\uff0c\u5c31\u53bb\u7db2\u8def\u4e0a\u722c\u87f2\uff0c\u7136\u5f8c\u914d\u5408\u9019\u500b\u65b9\u6cd5\u4f86\u5be6\u4f5c\u3002<\/li>
  6. Self-taught clustering\uff1a\u57fa\u790e\u65bc\u975e\u76e3\u7763\u5b78\u7fd2\u4e2d\u7684 Clustering \u5206\u7fa4(\u985e\u805a)\uff0c\u9019\u500b\u65b9\u6cd5\u7684\u8981\u6c42\u662f Source data \u4e00\u5b9a\u8981\u6bd4 Target data \u591a\u5f88\u591a\uff0c\u4e00\u4f46 Source data \u8db3\u4ee5\u751f\u6210\u6709\u6548\u7684 clusting \u6642\uff0c\u624d\u80fd\u628a\u9019\u500b\u6210\u679c\u79fb\u8f49\u5230 Target \u7136\u5f8c\u518d\u7528 unlabeled \u7684 data \u9032\u884c\u5206\u7fa4<\/li><\/ol>\n\n\n\n

    \u6311\u6230\u8207\u56de\u61c9\uff1a<\/strong><\/p>\n\n\n\n

    1. \u8981\u89e3\u6c7a\u7684\u554f\u984c\uff1a\u5728\u91ab\u5b78\u7684\u61c9\u7528\u4e2d\uff0c\u7f55\u898b\u75be\u75c5\u767c\u751f\u7684\u6a5f\u7387\u5c31\u4e0d\u9ad8\u4e86\uff0c\u8a31\u591a\u91ab\u5e2b\u4e5f\u662f\u5728\u5b78\u6821\u5b78\u904e\uff0c\u771f\u6b63\u5728\u81e8\u5e8a\u9047\u5230\u6642\u624d\u5728\u66f8\u4e0a\u3001\u7db2\u8def\u4e0a\u627e\u8cc7\u6599\u4f86\u5354\u52a9\u5224\u65b7\u7576\u524d\u7684\u75c5\u60a3\u662f\u5426\u78ba\u8a3a\uff1f\u540c\u7406\uff0c\u4f54\u53f0\u7063\u6700\u5927\u5b97\u7684\u88fd\u9020\u696d\uff0c\u5728\u505a\u826f\u7387\u6aa2\u9a57\u6642\uff0c\u5982\u679c\u9047\u4e0a\u4e86\u7455\u75b5\u7684 data \u91cf\u592a\u5c11\uff0c\u6216\u662f\u6b63\u5e38\u8207\u7455\u75b5\u7684\u91cf\u4e0d\u6210\u6bd4\u4f8b\uff0c\u4e0d\u662f\u900f\u904e Data augumentation(\u6b63\u898f\u5316\u3001\u767d\u5316\u3001\u5f71\u50cf\u8655\u7406) \u6216\u662f\u589e\/\u6e1b Data \u53ef\u4ee5\u89e3\u6c7a\u7684\u3002
      \u8a3b\uff1a\u6240\u8b02\u7684\u5f71\u50cf\u8655\u7406\uff0c\u6307\u7684\u662f\u653e\u5927\u3001\u7e2e\u5c0f\u3001\u88c1\u5207\u504f\u79fb\u3001\u7ffb\u8f49\uff0c\u751a\u81f3\u662f\u6539\u8b8a\u6bd4\u4f8b\u5c3a\u5bf8\u3001\u4eae\u5ea6\u8272\u6eab\u7b49\u2026\u624b\u6cd5<\/strong><\/li>
    2. \u76f8\u540c\u7684\u554f\u984c\uff1a\u5728\u50b3\u7d71\u7684ML\u4e2d\uff0c\u8a13\u7df4\u8207\u6e2c\u8a66\u8cc7\u6599\u96c6\u9700\u8981\u4f86\u81ea\u65bc\u76f8\u540c\u7684 feature space \u7279\u5fb5\u7a7a\u9593\uff0c\u63db\u53e5\u8a71\u8aaa\u4f60\u4e0d\u80fd\u62ff\u570b\u4e2d\u6578\u5b78\u53bb\u8a13\u7df4\u6a21\u578b\uff0c\u7136\u5f8c\u62ff\u9ad8\u4e2d\u6578\u5b78\u53bb\u6e2c\u8a66\u5b83\u3002\u56e0\u70ba\u9019\u4e8c\u500b\u8cc7\u6599\u96c6\u7684 Domain\u53ef\u80fd\u662f\u4e0d\u540c\u7684\uff0c\u8aaa\u7684\u66f4\u7cbe\u6e96\u4e9b\u662f feature space & probability \u53ef\u80fd\u6703\u4e00\u6a23\uff0c\u4f46\u4e5f\u53ef\u80fd\u662f\u4e0d\u4e00\u6a23\uff1b\u6216\u8005\u662f\u9019\u4e8c\u500b\u8cc7\u6599\u96c6\u7684 Task \u53ef\u80fd\u662f\u4e0d\u540c\u7684\uff0c\u8aaa\u7684\u66f4\u7cbe\u6e96\u4e9b\u662f label space \u6a19\u8a18\u7a7a\u9593 & object predictive function \u76ee\u6a19\u9810\u6e2c\u51fd\u5f0f \u53ef\u80fd\u6703\u4e00\u6a23\uff0c\u4f46\u4e5f\u53ef\u80fd\u662f\u4e0d\u4e00\u6a23\u3002\u540c\u7406\uff0c\u9664\u4e86\u6a21\u578b\u5728\u521d\u59cb\u751f\u6210\u6709\u9019\u6a23\u7684\u554f\u984c\u9700\u8981\u88ab\u8003\u616e\uff0c\u96a8\u8457\u6642\u7a7a\u63a8\u79fb\uff0c\u56e0\u70ba\u4e3b\u5ba2\u89c0\u74b0\u5883\u7684\u8b8a\u5316\uff0c\u5c07\u6703\u51fa\u73fe Model update \u7684\u72c0\u614b\u3002\u5c31\u50cf\u4e00\u4efd\u8003\u53e4\u984c\u4e0d\u53ef\u80fd\u6c38\u9060\u6709\u6548\uff0c\u9700\u8981\u88ab\u66f4\u65b0\u2026 <\/li>
    3. \u70ba\u4e86\u89e3\u6c7a\u4e0a\u8ff0\u7684\u554f\u984c\uff0c\u9077\u79fb\u5f0f\u5b78\u7fd2\u900f\u904e\u4e86Source & Target \u5728 instance \u6a23\u672c\u3001feature \u7279\u5fb5\u3001parameter \u53c3\u6578\u3001relation \u95dc\u4fc2\u2026\u7b49\u79fb\u8f49\u7684\u624b\u6cd5\u8b93 ML\u53ef\u4ee5\u8207\u6642\u4ff1\u9032\uff0c\u66f4\u52a0\u6210\u719f\u5730\u89e3\u6c7a\u66f4\u591a\u7684\u6311\u6230\u3002\u5c24\u5176\u662f\u9047\u5230\u6108\u4f86\u6108\u96e3\u53d6\u5f97\u500b\u8cc7\u7684\u73fe\u4ee3 Data set \u53d6\u5f97\u6311\u6230\uff0c\u6216\u662f Covid-19 \u5f71\u97ff\u4eba\u985e\u7684\u751f\u6d3b\u7fd2\u6163\u8b93\u4e4b\u524d\u5df2\u7d93\u751f\u6210\u597d\u7684\u6a21\u578b\uff0c\u9664\u4e86\u91cd\u65b0\u6536\u96c6 data \u4e26\u91cd\u65b0\u8a13\u7df4\u4e4b\u5916\uff0c\u4e5f\u6709\u5176\u4ed6\u6bd4\u8f03\u7701\u6642\u7701\u529b\u7684\u4f5c\u6cd5\u2026<\/li><\/ol>\n\n\n\n

      \u5be6\u4f5c\uff1a
      <\/strong>\u65e2\u7136\u662f\u521d\u9ad4\u9a57\uff0c\u6211\u5011\u5c31\u5f9e Fine tuning\u958b\u59cb\uff0c\u4ee5\u4e0b\u7684\u7bc4\u4f8b\u6211\u5c07\u63a1\u7528       Keras\u7684 Applications API<\/a>\uff0c\u4ee5\u8fd1300\u5f3515\u500b\u985e\u5225\u7684\u5efa\u7bc9\u7269\u8fa8\u8b58\u984c\u76ee\uff0c\u4f86\u505a\u9077\u79fb\u5f0f\u5b78\u7fd2\u7684 demo\u3002\u4f46\u7919\u65bc\u7bc7\u5e45\u6709\u9650\uff0c\u5c31\u4e0d\u518d\u8d05\u8ff0 Python code \u4e2d\u8f09\u5165\u51fd\u5f0f\u5eab\u3001\u8abf\u6574GPU\u53c3\u6578\u3001\u8b80\u53d6\u5716\u7247\u3001\u524d\u8655\u7406\uff0c\u76f4\u63a5\u5e6b\u5927\u5bb6\u5283\u91cd\u9ede\u5728 Transfer learning\u7684\u7d30\u7bc0(\u4e0b\u9762\u7684\u7406\u8ad6\u5716\u89e3\u6211\u662f\u53d6\u81ea\u4eba\u5de5\u667a\u6167\u5b78\u6821<\/a> AIA\u7684\u6587\u7ae0)<\/p>\n\n\n\n

      1. \u8f09\u5165 Pre-trained Model\uff1a <\/li><\/ol>\n\n\n\n

        \u4e0d\u7ba1\u5728\u4ec0\u9ebc\u6642\u4ee3\uff0c\u7ad9\u5728\u5de8\u4eba\u7684\u80a9\u8180\u4e00\u76f4\u90fd\u662f\u76f8\u7576\u76f4\u89c0\u4e14\u9ad8\u6548\u7684\u4f5c\u6cd5\uff0c2006\u5e74\u674e\u98db\u98db\u6559\u6388\u6240\u5275\u7acb\u7684 Imagenet\uff0c\u7d93\u904e\u591a\u5e74\u7684\u7d2f\u7a4d\uff0c\u5df2\u7d93\u8d85\u904e\u4e861400\u842c\u5f35\u7167\u7247(Multiple labeled)\uff0c\u88ab\u5206\u985e\u57282\u842c\u500b\u985e\u5225\uff0c\u5341\u591a\u5e74\u4f86\u9019\u500b Image recognition\u984c\u76ee\u7684\u6b66\u6797\u5927\u6703\uff0c\u5438\u5f15\u4e86\u591a\u5c11\u7684\u5168\u7403\u77e5\u540d\u7684\u5b78\u8005\u3001\u9ad8\u624b\uff0c\u4e00\u76f4\u5230\u79d1\u6280\u5de8\u64d8\u90fd\u8eac\u9022\u5176\u76db\uff0c\u8b93 AI\u5728\u5f71\u50cf\u8fa8\u8b58\u53ef\u4ee5\u843d\u5730\u5230\u6c11\u751f\u7684\u61c9\u7528(\u4f8b\u5982\uff0c\u5927\u7d04\u4e94\u5e74\u524d\u7576\u8fa8\u8b58\u7387\u9054\u5230 95%\u53f0\u7063\u7684\u5927\u90e8\u4efd\u505c\u8eca\u5834\u90fd\u958b\u59cb\u63a1\u7528\uff0c\u4f86\u7bc0\u7701\u4eba\u529b\uff0c\u52a0\u901f\u8eca\u8f1b\u901a\u904e\u7684\u6642\u9593)<\/p>\n\n\n\n

        \"Imagenet\"<\/figure>\n\n\n\n

        \u6211\u5011\u4f86\u770b\u4e00\u4e0b\u7a0b\u5f0f\u78bc\u7684\u90e8\u4efd\uff1a\u76ee\u524d Keras\u5df2\u7d93\u63d0\u4f9b\u4e8625\u500b\u6b66\u6797\u5927\u6703\u4e2d\u5f97\u540d\u7684\u6a21\u578b\uff0c\u6545\uff0c\u6211\u8f09\u5165\u4e86\u5165\u9580\u768416\u5c64\u7684 VGG16\uff0c\u4e26\u4e14\u900f\u904e Weights = ‘imagenet’ \u4f86\u5f97\u5230\u9ad8\u624b\u7684\u53c3\u6578\u503c\uff0cinclude_top = false \u5728\u6a21\u578b\u7684\u6700\u5de6\u908a(\u5168\u9023\u63a5\u5c64)\u4e2d\uff0c\u7576\u6211\u8981\u63db\u6210\u81ea\u5df1\u7684\u5716\u7247\uff0c\u5c31\u8981\u9019\u6a23\u8a2d\u5b9a\u3002\u63a5\u8457\uff0c\u5c31\u662f\u6a19\u6e96\u7684 CNN \u6a21\u578b\u8a2d\u5b9a\uff0c\u7d30\u7bc0\u53ef\u4ee5\u53c3\u8003\u4e4b\u524d\u7684\u6587\u7ae0\u3002\u7528 summary() \u9810\u89bd\u4e00\u4e0b\u6a21\u578b\u7684\u6a23\u8c8c\u8207\u53c3\u6578\u6578\u91cf
        \u5176\u4e2d\u6709\u4e00\u500b Flatten() \u662f\u5728\u5efaCCN\u6a21\u578b\u4e2d\uff0c\u5e38\u7528\u65bc\u5c3e\u8072\u8981\u628a Convolution \u5377\u7a4d\u5c64\u518d\u63a5\u5230\u5168\u9023\u63a5\u5c64\uff0c\u8981\u505a Softmax \u9810\u6e2c\u6642\u7684\u4e00\u500b\u624b\u6cd5\u904e\u7a0b\u2026 <\/p>\n\n\n\n

        \"Transfer<\/figure>\n\n\n\n
        \"Transfer<\/figure>\n\n\n\n
        1. \u6c7a\u5b9a Pre-Train \u53c3\u6578\u91cf\uff1a<\/strong><\/li><\/ol>\n\n\n\n

          \u6211\u5011\u8981\u6372\u8d77\u8896\u5b50\u9032\u884c\u51cd\u7d50\u53c3\u6578\u7684\u52d5\u4f5c\uff0c\u5728\u958b\u59cb\u4e4b\u524d\uff0c\u8981\u5148\u77e5\u9053\u4e00\u500b\u7a69\u5b9a\u7684\u6a21\u578b\u5176 Learning rate\u5df2\u7d93\u9054\u5230\u6700\u4f73\u5316\u4e86\uff0c\u91cd\u65b0\u8a13\u7df4\u6642\u6703\u6709 over fiting\u7684\u98a8\u96aa\u3002\u5728\u4e0b\u5716\u4e2d\uff0c\u4f60\u53ef\u4ee5\u770b\u5230CNN\u5728\u4f4e\u968e\u5c64(\u5de6\u908a)\u6642\uff0c\u9084\u5728\u5b78\u7fd2\u9ede\u3001\u7dda\u3001\u908a\u7de3\u2026\u9019\u500b\u5b30\u5152\u7d1a\u7684\u5e38\u8b58\uff0c\u5230\u4e86\u9ad8\u968e\u5c64(\u53f3\u908a)\u6642\uff0c\u624d\u80fd\u5b78\u7fd2\u8a31\u591a\u5c0f\u7269\u4ef6\u4f75\u6e4a\u51fa\u4f86\u7684\u7684\u751f\u7269\u6216\u7269\u9ad4\u3002
          \u4f9d\u64da Source\/ Target \u8cc7\u6599\u96c6\u7684\u76f8\u4f3c\u5ea6\uff0c\u4f60\u6709\u4ee5\u4e0b\u7684\u9078\u64c7\uff1a
          A. \u5982\u679c\u4f60\u7684\u8cc7\u6599\u96c6\u4e26\u4e0d\u5927\uff0c\u7576\u4e8c\u500b\u8cc7\u6599\u96c6\u6bd4\u8f03\u76f8\u4f3c\u6642\uff0c\u53ea\u9700\u9077\u79fb\u9ad8\u968e\u5c64\u7684\u53c3\u6578\uff0c\u4ee5\u5fae\u8abf\u7684\u65b9\u5f0f\uff0c\u8b93 Learning rate \u63a7\u5236\u5728\u6bd4\u8f03\u5c0f\uff1b\u540c\u7406\uff0c\u7576\u4e8c\u500b\u8cc7\u6599\u96c6\u6bd4\u8f03\u4e0d\u76f8\u4f3c\u6642\uff0c\u4e5f\u53ea\u9700\u8981\u9077\u79fb\u4f4e\u968e\u5c64\u7684\u53c3\u6578\uff0c\u5373\u53ef\u3002
          B. \u5982\u679c\u4f60\u7684\u8cc7\u6599\u96c6\u5920\u5927\uff0c\u5982\u679c\u4e8c\u500b\u8cc7\u6599\u96c6\u6bd4\u8f03\u76f8\u4f3c\u6642\uff0c\u4f60\u53ef\u4ee5\u5c07 Pre-Train \u6a21\u578b\u7684\u6b0a\u91cd\u7576\u4f5c\u8d77\u59cb\u503c\uff0c\u5168\u90e8\u518d\u8a13\u7df4\u904e\u4e00\u6b21\uff0c\u53ea\u662f Learning rate \u6bd4\u8f03\u5927\uff1b\u5982\u679c\u4e8c\u500b\u8cc7\u6599\u96c6\u6bd4\u8f03\u4e0d\u76f8\u4f3c\u6642\uff0c\u4f60\u53ef\u4ee5\u7565\u904e Pre-Train model \u81ea\u5df1\u4f86<\/p>\n\n\n\n

          \"pre-train-model\"<\/figure>\n\n\n\n

          \u4ee5\u672c\u6587\u7684\u6848\u4f8b\u4f86\u8aaa\uff0c\u7531\u65bc Source\/Target \u7684\u8cc7\u6599\u96c6\u76f8\u4f3c\uff0c\u6240\u4ee5\u6211\u5c31\u51cd\u7d50\u5927\u90e8\u4efd\u7684\u6a21\u578b\uff0c\u53ea\u91cd\u65b0\u8a13\u7df4\u9ad8\u968e\u5c64\u7684 3\u5c64\uff0c\u5148\u8a2d\u5b9a trainable = True\uff0ctrainable_layer = 3\uff0c\u7136\u5f8c\u518d\u7528\u4e00\u500b\u5efb\u5708\u5c07\u5176\u4ed6\u968e\u5c64\u51cd\u7d50\u8d77\u4f86\u3002\u6700\u5f8c\uff0c\u518d\u5370\u51fa\u4f86\u6aa2\u67e5\u4e00\u4e0b\u2026\u7136\u5f8c\u628a\u5176\u4ed6\u6a21\u578b\u8a13\u7df4\u7684\u8d85\u53c3\u6578\u8a2d\u5b9a\u597d\uff0c\u5c31\u6e96\u5099\u8981\u958b\u59cb\u8a13\u7df4\u4e86<\/p>\n\n\n\n

          \"Transfer<\/figure>\n\n\n\n
          \"Transfer<\/figure>\n\n\n\n
          1. \u958b\u59cb\u8a13\u7df4\u6a21\u578b\uff1a<\/strong><\/li><\/ol>\n\n\n\n

            \u6709\u4e86\u9077\u79fb\u5f0f\u5b78\u7fd2\uff0c\u53ea\u8981\u7d93\u904e10\u500b Epoches \u6b63\u78ba\u7387\u5c31\u53ef\u4ee5\u9054\u5230 95%<\/p>\n\n\n\n

            \"Transfer<\/figure>\n\n\n\n
            \"Transfer<\/figure>\n\n\n\n
            1. \u8996\u89ba\u5316\u8a13\u7df4\u7684\u7d50\u679c\uff1a<\/li><\/ol>\n\n\n\n

              \u63a5\u8457\u6211\u5011\u5c31\u62ff\u51fa\u8fd1300\u7b46\u7684\u6e2c\u8a66\u8cc7\u6599\u4f86\u9a57\u8b49\u6211\u5011\u7684\u6a21\u578b\u5427\uff01<\/p>\n\n\n\n

              \"Transfer<\/figure>\n\n\n\n

              \u756b\u51fa\u8a13\u7df4\u8207\u6e2c\u8a66\u968e\u6bb5\u7684\u6bd4\u8f03\u6298\u7dda\u5716<\/p>\n\n\n\n

              \"modelval-loss\"<\/figure>\n\n\n\n

              \u63a5\u8457\uff0c\u518d\u900f\u904e\u50b3\u7d71\u8207 Seaborn\u7684\u6df7\u6dc6\u77e9\u9663\uff0c\u4e86\u89e3\u6e2c\u8a66\u6642\u7684\u932f\u8aa4\u5206\u4f48<\/p>\n\n\n\n

              \"Transfer<\/figure>\n\n\n\n
              \"\"<\/figure>\n\n\n\n

              Seaborn \u7684\u8996\u89ba\u5316\u5448\u73fe\u6548\u679c\u597d\uff0c\u9069\u5408\u62ff\u4f86\u505a\u5831\u544a<\/p>\n\n\n\n

              \"confusion-matrix\"<\/figure>\n\n\n\n

              \u672a\u4f86\u5c55\u671b\uff1a
              <\/strong>\u9664\u4e86 Transfer learning\u4e4b\u5916\uff0c\u9084\u6709\u4e00\u500b Knowledge distillation \u53ef\u4ee5\u5c07\u5177\u6709\u773e\u591a\u53c3\u6578\u7684\u5927\u578b AI \u6a21\u578b\uff0c\u62bd\u51fa\u7cbe\u83ef\uff0c\u5c0d\u5c0f\u578b\u7684 AI\u6a21\u578b\u505a\u5178\u7bc4\u79fb\u8f49\u3002\u9019\u88e1\u6709\u4e8c\u500b\u5c64\u9762\u9700\u8981\u8003\u91cf\uff0c\u4e00\u500b\u662f AI\u5982\u4f55\u843d\u5730\uff1f\u4e00\u500b\u662f\u5982\u4f55\u56e0\u624d\u65bd\u6559\uff1f\u524d\u8005\u662f\uff0c\u4e0d\u53ef\u80fd\u96a8\u6642\u80fd\u5920\u642c\u4e00\u500b GPU \u6a5f\u6ac3\u5728\u8def\u4e0a\u8d70\uff0c\u6216\u662f\u4f60\u7684\u7522\u54c1\u5167\u5efa\u6709 Wifi \u6216\u662f 5G\u884c\u52d5\u4e0a\u7db2\u80fd\u529b\uff0c\u53ef\u4ee5\u9023\u7d50\u5230\u96f2\u7aef\u5927\u578b\u7684 AI\u6a21\u578b\u3002\u5f8c\u8005\u662f\uff0c\u7576\u4f60\u7684\u5152\u5b50\u7d42\u65bc\u904e\u4e86\u5f37\u8913\u968e\u6bb5\u8981\u4e0a\u5e7c\u7a1a\u5712\uff0c\u5728\u4ed6\u4eba\u751f\u7b2c\u4e00\u6b21\u7684\u4e0a\u5b78\u7684\u7d93\u5178\u65b7\u6368\u96e2\u5834\u666f\u4e2d\uff0c\u4f60\u627e\u4e00\u500b\u5927\u5b78\u6559\u6388\u4f86\u6559\u80b2\u4f60\u7684\u5bf6\u8c9d\uff0c\u7531\u65bc\u6559\u6388\u4e26\u4e0d\u64c5\u9577\u8655\u7406\u4e00\u54ed\u4e8c\u9b27\u4e09\u4e0a\u540a\u7684\u5c0f\u5c0f\u5b69\u53cd\u61c9\u3002\u9084\u4e0d\u5982\u627e\u4e00\u500b\u4e09\u5341\u521d\u982d\u7684\u5e7c\u7a1a\u5712\u4e3b\u4efb\uff0c\u4f86\u7684\u9069\u5408\u3002\u56e0\u70ba\u50cf\u662f GPT-2 \u9019\u7a2e\u64c1\u6709\u4e0a\u5104\u53c3\u6578\u7b49\u7d1a\u7684\u7368\u89d2\u7378\u5927\u6a21\u578b\uff0c\u4f60\u662f\u5f88\u96e3\u76f4\u63a5\u653e\u5728\u4f60\u7684\u624b\u6a5f\u4e2d\uff0c\u6703\u9700\u8981\u50cf\u662f\u77e5\u8b58\u84b8\u993e\u9019\u985e\u7684\u624b\u6cd5\uff0c\u624d\u80fd\u6210\u529f\u843d\u5730\u7684\u3002\u5c07\u4f86\u6709\u6a5f\u6703\u518d\u4f86\u4ecb\u7d39\u7d66\u5404\u4f4d\u2026<\/p>\n\n\n\n

              \u767c\u4f48\u65e5\u671f : 2021-03-04
              \u672c\u6587\u7d93\u5408\u4f5c\u5925\u4f34 \u6388\u6b0a\u8f49\u8f09\u81ea <\/a>Transfer learning \u521d\u9ad4\u9a57<\/a> (Christian Lee)<\/p>","protected":false},"excerpt":{"rendered":"

              \u4e0d\u66c9\u5f97\u4f60\u6709\u6c92\u6709\u60f3\u904e\u6bd4\u8d77\u5176\u4ed6\u98df\u7269\u93c8\u9802\u7aef\u7684\u5176\u4ed6\u751f\u7269\uff0c\u70ba\u4f55\u53ea\u6709\u4eba\u985e\u53ef\u4ee5\u767c\u5c55\u9ad8\u5ea6\u6587\u660e\uff0c\u751a\u81f3\u53ef\u4ee5\u822a\u5411\u5b87\u5b99\u661f\u8fb0\uff1f\u7b54\u6848\u5c31\u662f\u5178\u7bc4\u79fb\u8f49\uff01\u672c\u6587\u5c07\u6703\u4ecb\u7d39\u5728AI\u7684\u4e16\u754c\u4e86\u5982\u4f55\u61c9\u7528\u9019\u500b\u6280\u8853\uff0c\u7528\u6700\u77ed\u7684\u6642\u9593\u8207\u6700\u5c0f\u7684\u8cc7\u6599\u96c6\uff0c\u8a13\u7df4\u51fa\u4e0d\u932f\u7684AI\u6a21\u578b\u2026<\/p>","protected":false},"author":7,"featured_media":18893,"menu_order":58,"comment_status":"closed","ping_status":"closed","template":"","format":"standard","meta":{"_acf_changed":false,"inline_featured_image":false,"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"sector":[],"solutions":[7190,7196],"brand":[7167],"class_list":["post-78010","products","type-products","status-publish","format-standard","has-post-thumbnail","hentry","solutions-cloud-architecture-optimization","solutions-value-added-services","brand-microsoft-azure"],"acf":[],"yoast_head":"\nTransfer learning \u521d\u9ad4\u9a57 - WingWill \u7fbd\u6607\u570b\u969b - \u5168\u65b9\u4f4d\u96f2\u7aef\u3001\u8cc7\u5b89\u670d\u52d9\uff0c\u6578\u4f4d\u8f49\u578b\u597d\u5925\u4f34<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wingwill.com.tw\/ja\/products\/blog-a61-transfer-learning\/\" \/>\n<meta property=\"og:locale\" content=\"ja_JP\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"Transfer learning \u521d\u9ad4\u9a57 - WingWill \u7fbd\u6607\u570b\u969b - \u5168\u65b9\u4f4d\u96f2\u7aef\u3001\u8cc7\u5b89\u670d\u52d9\uff0c\u6578\u4f4d\u8f49\u578b\u597d\u5925\u4f34\" \/>\n<meta property=\"og:description\" content=\"\u4e0d\u66c9\u5f97\u4f60\u6709\u6c92\u6709\u60f3\u904e\u6bd4\u8d77\u5176\u4ed6\u98df\u7269\u93c8\u9802\u7aef\u7684\u5176\u4ed6\u751f\u7269\uff0c\u70ba\u4f55\u53ea\u6709\u4eba\u985e\u53ef\u4ee5\u767c\u5c55\u9ad8\u5ea6\u6587\u660e\uff0c\u751a\u81f3\u53ef\u4ee5\u822a\u5411\u5b87\u5b99\u661f\u8fb0\uff1f\u7b54\u6848\u5c31\u662f\u5178\u7bc4\u79fb\u8f49\uff01\u672c\u6587\u5c07\u6703\u4ecb\u7d39\u5728AI\u7684\u4e16\u754c\u4e86\u5982\u4f55\u61c9\u7528\u9019\u500b\u6280\u8853\uff0c\u7528\u6700\u77ed\u7684\u6642\u9593\u8207\u6700\u5c0f\u7684\u8cc7\u6599\u96c6\uff0c\u8a13\u7df4\u51fa\u4e0d\u932f\u7684AI\u6a21\u578b\u2026\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wingwill.com.tw\/ja\/products\/blog-a61-transfer-learning\/\" \/>\n<meta property=\"og:site_name\" content=\"WingWill \u7fbd\u6607\u570b\u969b - \u5168\u65b9\u4f4d\u96f2\u7aef\u3001\u8cc7\u5b89\u670d\u52d9\uff0c\u6578\u4f4d\u8f49\u578b\u597d\u5925\u4f34\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/WingWillCo.ltd\/\" \/>\n<meta property=\"article:modified_time\" content=\"2025-11-27T01:38:00+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wingwill.com.tw\/wp-content\/uploads\/2021\/03\/selected-blog-big-data-and-machinelearning-3.jpg\" \/>\n\t<meta property=\"og:image:width\" content=\"1900\" \/>\n\t<meta property=\"og:image:height\" content=\"1080\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/jpeg\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u63a8\u5b9a\u8aad\u307f\u53d6\u308a\u6642\u9593\" \/>\n\t<meta name=\"twitter:data1\" content=\"2\u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":[\"Article\",\"BlogPosting\"],\"@id\":\"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/#article\",\"isPartOf\":{\"@id\":\"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/\"},\"author\":{\"name\":\"Chen Enfys\",\"@id\":\"https:\/\/www.wingwill.com.tw\/#\/schema\/person\/b9809ad749ffc04c733fa34746bc1440\"},\"headline\":\"Transfer learning \u521d\u9ad4\u9a57\",\"datePublished\":\"2021-03-05T07:01:00+00:00\",\"dateModified\":\"2025-11-27T01:38:00+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/\"},\"wordCount\":346,\"publisher\":{\"@id\":\"https:\/\/www.wingwill.com.tw\/#organization\"},\"image\":{\"@id\":\"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/www.wingwill.com.tw\/wp-content\/uploads\/2021\/03\/selected-blog-big-data-and-machinelearning-3.jpg\",\"keywords\":[\"Azure\",\"Microsoft\",\"Machine Learning\",\"Algorithm\",\"DNN\",\"\u5fae\u8edf\"],\"articleSection\":[\"\u5927\u6578\u64da\u8207\u6a5f\u5668\u5b78\u7fd2\"],\"inLanguage\":\"ja\"},{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/\",\"url\":\"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/\",\"name\":\"Transfer learning \u521d\u9ad4\u9a57 - WingWill \u7fbd\u6607\u570b\u969b - \u5168\u65b9\u4f4d\u96f2\u7aef\u3001\u8cc7\u5b89\u670d\u52d9\uff0c\u6578\u4f4d\u8f49\u578b\u597d\u5925\u4f34\",\"isPartOf\":{\"@id\":\"https:\/\/www.wingwill.com.tw\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/#primaryimage\"},\"image\":{\"@id\":\"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/#primaryimage\"},\"thumbnailUrl\":\"https:\/\/www.wingwill.com.tw\/wp-content\/uploads\/2021\/03\/selected-blog-big-data-and-machinelearning-3.jpg\",\"datePublished\":\"2021-03-05T07:01:00+00:00\",\"dateModified\":\"2025-11-27T01:38:00+00:00\",\"breadcrumb\":{\"@id\":\"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/#breadcrumb\"},\"inLanguage\":\"ja\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"ja\",\"@id\":\"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/#primaryimage\",\"url\":\"https:\/\/www.wingwill.com.tw\/wp-content\/uploads\/2021\/03\/selected-blog-big-data-and-machinelearning-3.jpg\",\"contentUrl\":\"https:\/\/www.wingwill.com.tw\/wp-content\/uploads\/2021\/03\/selected-blog-big-data-and-machinelearning-3.jpg\",\"width\":1900,\"height\":1080},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9801\",\"item\":\"https:\/\/www.wingwill.com.tw\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"\u89e3\u6c7a\u65b9\u6848\",\"item\":\"https:\/\/www.wingwill.com.tw\/products\/\"},{\"@type\":\"ListItem\",\"position\":3,\"name\":\"Transfer learning \u521d\u9ad4\u9a57\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wingwill.com.tw\/#website\",\"url\":\"https:\/\/www.wingwill.com.tw\/\",\"name\":\"WingWill \u7fbd\u6607\u570b\u969b - \u5168\u65b9\u4f4d\u96f2\u7aef\u3001\u8cc7\u5b89\u670d\u52d9\uff0c\u6578\u4f4d\u8f49\u578b\u597d\u5925\u4f34\",\"description\":\"Cloud Solution Provider\uff5c\u7fbd\u6607\u570b\u969b\u70ba\u96f6\u58f9\u79d1\u6280\u767e\u5206\u4e4b\u767e\u6295\u8cc7\u7684\u96f2\u7aef\u670d\u52d9\u4f9b\u61c9\u5546\uff0c\u4ee5\u5c08\u696d\u6280\u8853\u8207\u7d93\u9a57\u6df1\u8015\u96f2\u7aef\u5e02\u5834\uff0c\u70baGoogle Cloud\u3001Akamai\u3001\u5fae\u8edf\u3001AWS \u7b49\u5404\u570b\u969b\u5927\u5ee0\u4e4b\u5408\u4f5c\u5925\u4f34\uff0c\u63d0\u4f9b\u61c9\u7528\u670d\u52d9\u50b3\u905e\u3001\u96f2\u5730\u6df7\u5408\u57fa\u790e\u8a2d\u8a08\u3001\u96f2\u7aef\u5b89\u5168\u7ba1\u7406\u3001\u5927\u6578\u64da\u5206\u6790\u7b49\u56db\u5927\u670d\u52d9\uff0c\u5354\u52a9\u4f01\u696d\u52a0\u901f\u6578\u4f4d\u8f49\u578b\uff0c\u63d0\u4f9b\u6700\u4f73\u7684\u96f2\u7aef\u9632\u8b77\u89e3\u6c7a\u65b9\u6848\u3002\",\"publisher\":{\"@id\":\"https:\/\/www.wingwill.com.tw\/#organization\"},\"alternateName\":\"WingWill \u7fbd\u6607\u570b\u969b - \u5168\u65b9\u4f4d\u96f2\u7aef\u3001\u8cc7\u5b89\u670d\u52d9\uff0c\u6578\u4f4d\u8f49\u578b\u597d\u5925\u4f34\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wingwill.com.tw\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"ja\"},{\"@type\":\"Organization\",\"@id\":\"https:\/\/www.wingwill.com.tw\/#organization\",\"name\":\"WingWill \u7fbd\u6607\u570b\u969b\",\"alternateName\":\"WingWill \u7fbd\u6607\u570b\u969b\",\"url\":\"https:\/\/www.wingwill.com.tw\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"ja\",\"@id\":\"https:\/\/www.wingwill.com.tw\/#\/schema\/logo\/image\/\",\"url\":\"https:\/\/www.wingwill.com.tw\/wp-content\/uploads\/2025\/11\/logo.png\",\"contentUrl\":\"https:\/\/www.wingwill.com.tw\/wp-content\/uploads\/2025\/11\/logo.png\",\"width\":955,\"height\":955,\"caption\":\"WingWill \u7fbd\u6607\u570b\u969b\"},\"image\":{\"@id\":\"https:\/\/www.wingwill.com.tw\/#\/schema\/logo\/image\/\"},\"sameAs\":[\"https:\/\/www.facebook.com\/WingWillCo.ltd\/\"],\"email\":\"service@wingwill.com.tw\",\"telephone\":\"02 2656 5630\",\"legalName\":\"\u7fbd\u6607\u570b\u969b\u80a1\u4efd\u6709\u9650\u516c\u53f8\",\"taxID\":\"52811521\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wingwill.com.tw\/#\/schema\/person\/b9809ad749ffc04c733fa34746bc1440\",\"name\":\"Chen Enfys\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"ja\",\"@id\":\"https:\/\/www.wingwill.com.tw\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/secure.gravatar.com\/avatar\/8dd4805ae5a55e6c0ceb3adb825b9cb840c911f0540b0ebf28091093aa70fb88?s=96&d=mm&r=g\",\"contentUrl\":\"https:\/\/secure.gravatar.com\/avatar\/8dd4805ae5a55e6c0ceb3adb825b9cb840c911f0540b0ebf28091093aa70fb88?s=96&d=mm&r=g\",\"caption\":\"Chen Enfys\"},\"url\":\"https:\/\/www.wingwill.com.tw\/ja\/author\/enfys-chenzerone-com-tw\/\"}]}<\/script>\n<!-- \/ Yoast SEO Premium plugin. -->","yoast_head_json":{"title":"Transfer learning \u521d\u9ad4\u9a57 - WingWill \u7fbd\u6607\u570b\u969b - \u5168\u65b9\u4f4d\u96f2\u7aef\u3001\u8cc7\u5b89\u670d\u52d9\uff0c\u6578\u4f4d\u8f49\u578b\u597d\u5925\u4f34","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wingwill.com.tw\/ja\/products\/blog-a61-transfer-learning\/","og_locale":"ja_JP","og_type":"article","og_title":"Transfer learning \u521d\u9ad4\u9a57 - WingWill \u7fbd\u6607\u570b\u969b - \u5168\u65b9\u4f4d\u96f2\u7aef\u3001\u8cc7\u5b89\u670d\u52d9\uff0c\u6578\u4f4d\u8f49\u578b\u597d\u5925\u4f34","og_description":"\u4e0d\u66c9\u5f97\u4f60\u6709\u6c92\u6709\u60f3\u904e\u6bd4\u8d77\u5176\u4ed6\u98df\u7269\u93c8\u9802\u7aef\u7684\u5176\u4ed6\u751f\u7269\uff0c\u70ba\u4f55\u53ea\u6709\u4eba\u985e\u53ef\u4ee5\u767c\u5c55\u9ad8\u5ea6\u6587\u660e\uff0c\u751a\u81f3\u53ef\u4ee5\u822a\u5411\u5b87\u5b99\u661f\u8fb0\uff1f\u7b54\u6848\u5c31\u662f\u5178\u7bc4\u79fb\u8f49\uff01\u672c\u6587\u5c07\u6703\u4ecb\u7d39\u5728AI\u7684\u4e16\u754c\u4e86\u5982\u4f55\u61c9\u7528\u9019\u500b\u6280\u8853\uff0c\u7528\u6700\u77ed\u7684\u6642\u9593\u8207\u6700\u5c0f\u7684\u8cc7\u6599\u96c6\uff0c\u8a13\u7df4\u51fa\u4e0d\u932f\u7684AI\u6a21\u578b\u2026","og_url":"https:\/\/www.wingwill.com.tw\/ja\/products\/blog-a61-transfer-learning\/","og_site_name":"WingWill \u7fbd\u6607\u570b\u969b - \u5168\u65b9\u4f4d\u96f2\u7aef\u3001\u8cc7\u5b89\u670d\u52d9\uff0c\u6578\u4f4d\u8f49\u578b\u597d\u5925\u4f34","article_publisher":"https:\/\/www.facebook.com\/WingWillCo.ltd\/","article_modified_time":"2025-11-27T01:38:00+00:00","og_image":[{"width":1900,"height":1080,"url":"https:\/\/www.wingwill.com.tw\/wp-content\/uploads\/2021\/03\/selected-blog-big-data-and-machinelearning-3.jpg","type":"image\/jpeg"}],"twitter_card":"summary_large_image","twitter_misc":{"\u63a8\u5b9a\u8aad\u307f\u53d6\u308a\u6642\u9593":"2\u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":["Article","BlogPosting"],"@id":"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/#article","isPartOf":{"@id":"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/"},"author":{"name":"Chen Enfys","@id":"https:\/\/www.wingwill.com.tw\/#\/schema\/person\/b9809ad749ffc04c733fa34746bc1440"},"headline":"Transfer learning \u521d\u9ad4\u9a57","datePublished":"2021-03-05T07:01:00+00:00","dateModified":"2025-11-27T01:38:00+00:00","mainEntityOfPage":{"@id":"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/"},"wordCount":346,"publisher":{"@id":"https:\/\/www.wingwill.com.tw\/#organization"},"image":{"@id":"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/#primaryimage"},"thumbnailUrl":"https:\/\/www.wingwill.com.tw\/wp-content\/uploads\/2021\/03\/selected-blog-big-data-and-machinelearning-3.jpg","keywords":["Azure","Microsoft","Machine Learning","Algorithm","DNN","\u5fae\u8edf"],"articleSection":["\u5927\u6578\u64da\u8207\u6a5f\u5668\u5b78\u7fd2"],"inLanguage":"ja"},{"@type":"WebPage","@id":"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/","url":"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/","name":"Transfer learning \u521d\u9ad4\u9a57 - WingWill \u7fbd\u6607\u570b\u969b - \u5168\u65b9\u4f4d\u96f2\u7aef\u3001\u8cc7\u5b89\u670d\u52d9\uff0c\u6578\u4f4d\u8f49\u578b\u597d\u5925\u4f34","isPartOf":{"@id":"https:\/\/www.wingwill.com.tw\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/#primaryimage"},"image":{"@id":"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/#primaryimage"},"thumbnailUrl":"https:\/\/www.wingwill.com.tw\/wp-content\/uploads\/2021\/03\/selected-blog-big-data-and-machinelearning-3.jpg","datePublished":"2021-03-05T07:01:00+00:00","dateModified":"2025-11-27T01:38:00+00:00","breadcrumb":{"@id":"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/#breadcrumb"},"inLanguage":"ja","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/"]}]},{"@type":"ImageObject","inLanguage":"ja","@id":"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/#primaryimage","url":"https:\/\/www.wingwill.com.tw\/wp-content\/uploads\/2021\/03\/selected-blog-big-data-and-machinelearning-3.jpg","contentUrl":"https:\/\/www.wingwill.com.tw\/wp-content\/uploads\/2021\/03\/selected-blog-big-data-and-machinelearning-3.jpg","width":1900,"height":1080},{"@type":"BreadcrumbList","@id":"https:\/\/www.wingwill.com.tw\/en\/products\/blog-a61-transfer-learning\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9801","item":"https:\/\/www.wingwill.com.tw\/"},{"@type":"ListItem","position":2,"name":"\u89e3\u6c7a\u65b9\u6848","item":"https:\/\/www.wingwill.com.tw\/products\/"},{"@type":"ListItem","position":3,"name":"Transfer learning \u521d\u9ad4\u9a57"}]},{"@type":"WebSite","@id":"https:\/\/www.wingwill.com.tw\/#website","url":"https:\/\/www.wingwill.com.tw\/","name":"WingWill \u7fbd\u6607\u570b\u969b - \u5168\u65b9\u4f4d\u96f2\u7aef\u3001\u8cc7\u5b89\u670d\u52d9\uff0c\u6578\u4f4d\u8f49\u578b\u597d\u5925\u4f34","description":"Cloud Solution Provider\uff5c\u7fbd\u6607\u570b\u969b\u70ba\u96f6\u58f9\u79d1\u6280\u767e\u5206\u4e4b\u767e\u6295\u8cc7\u7684\u96f2\u7aef\u670d\u52d9\u4f9b\u61c9\u5546\uff0c\u4ee5\u5c08\u696d\u6280\u8853\u8207\u7d93\u9a57\u6df1\u8015\u96f2\u7aef\u5e02\u5834\uff0c\u70baGoogle Cloud\u3001Akamai\u3001\u5fae\u8edf\u3001AWS \u7b49\u5404\u570b\u969b\u5927\u5ee0\u4e4b\u5408\u4f5c\u5925\u4f34\uff0c\u63d0\u4f9b\u61c9\u7528\u670d\u52d9\u50b3\u905e\u3001\u96f2\u5730\u6df7\u5408\u57fa\u790e\u8a2d\u8a08\u3001\u96f2\u7aef\u5b89\u5168\u7ba1\u7406\u3001\u5927\u6578\u64da\u5206\u6790\u7b49\u56db\u5927\u670d\u52d9\uff0c\u5354\u52a9\u4f01\u696d\u52a0\u901f\u6578\u4f4d\u8f49\u578b\uff0c\u63d0\u4f9b\u6700\u4f73\u7684\u96f2\u7aef\u9632\u8b77\u89e3\u6c7a\u65b9\u6848\u3002","publisher":{"@id":"https:\/\/www.wingwill.com.tw\/#organization"},"alternateName":"WingWill \u7fbd\u6607\u570b\u969b - \u5168\u65b9\u4f4d\u96f2\u7aef\u3001\u8cc7\u5b89\u670d\u52d9\uff0c\u6578\u4f4d\u8f49\u578b\u597d\u5925\u4f34","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wingwill.com.tw\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"ja"},{"@type":"Organization","@id":"https:\/\/www.wingwill.com.tw\/#organization","name":"WingWill \u7fbd\u6607\u570b\u969b","alternateName":"WingWill \u7fbd\u6607\u570b\u969b","url":"https:\/\/www.wingwill.com.tw\/","logo":{"@type":"ImageObject","inLanguage":"ja","@id":"https:\/\/www.wingwill.com.tw\/#\/schema\/logo\/image\/","url":"https:\/\/www.wingwill.com.tw\/wp-content\/uploads\/2025\/11\/logo.png","contentUrl":"https:\/\/www.wingwill.com.tw\/wp-content\/uploads\/2025\/11\/logo.png","width":955,"height":955,"caption":"WingWill \u7fbd\u6607\u570b\u969b"},"image":{"@id":"https:\/\/www.wingwill.com.tw\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/WingWillCo.ltd\/"],"email":"service@wingwill.com.tw","telephone":"02 2656 5630","legalName":"\u7fbd\u6607\u570b\u969b\u80a1\u4efd\u6709\u9650\u516c\u53f8","taxID":"52811521"},{"@type":"Person","@id":"https:\/\/www.wingwill.com.tw\/#\/schema\/person\/b9809ad749ffc04c733fa34746bc1440","name":"Chen Enfys","image":{"@type":"ImageObject","inLanguage":"ja","@id":"https:\/\/www.wingwill.com.tw\/#\/schema\/person\/image\/","url":"https:\/\/secure.gravatar.com\/avatar\/8dd4805ae5a55e6c0ceb3adb825b9cb840c911f0540b0ebf28091093aa70fb88?s=96&d=mm&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/8dd4805ae5a55e6c0ceb3adb825b9cb840c911f0540b0ebf28091093aa70fb88?s=96&d=mm&r=g","caption":"Chen Enfys"},"url":"https:\/\/www.wingwill.com.tw\/ja\/author\/enfys-chenzerone-com-tw\/"}]}},"_links":{"self":[{"href":"https:\/\/www.wingwill.com.tw\/ja\/wp-json\/wp\/v2\/products\/78010","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wingwill.com.tw\/ja\/wp-json\/wp\/v2\/products"}],"about":[{"href":"https:\/\/www.wingwill.com.tw\/ja\/wp-json\/wp\/v2\/types\/products"}],"author":[{"embeddable":true,"href":"https:\/\/www.wingwill.com.tw\/ja\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wingwill.com.tw\/ja\/wp-json\/wp\/v2\/comments?post=78010"}],"version-history":[{"count":0,"href":"https:\/\/www.wingwill.com.tw\/ja\/wp-json\/wp\/v2\/products\/78010\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wingwill.com.tw\/ja\/wp-json\/wp\/v2\/media\/18893"}],"wp:attachment":[{"href":"https:\/\/www.wingwill.com.tw\/ja\/wp-json\/wp\/v2\/media?parent=78010"}],"wp:term":[{"taxonomy":"sector","embeddable":true,"href":"https:\/\/www.wingwill.com.tw\/ja\/wp-json\/wp\/v2\/sector?post=78010"},{"taxonomy":"solutions","embeddable":true,"href":"https:\/\/www.wingwill.com.tw\/ja\/wp-json\/wp\/v2\/solutions?post=78010"},{"taxonomy":"brand","embeddable":true,"href":"https:\/\/www.wingwill.com.tw\/ja\/wp-json\/wp\/v2\/brand?post=78010"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}