{"id":477793,"date":"2023-08-09T09:20:26","date_gmt":"2023-08-09T09:20:26","guid":{"rendered":""},"modified":"2023-09-05T11:15:25","modified_gmt":"2023-09-05T11:15:25","slug":"label-smoothing","status":"publish","type":"wiki","link":"https:\/\/oneproxy.pro\/cn\/wiki\/label-smoothing\/","title":{"rendered":"\u6807\u7b7e\u5e73\u6ed1"},"content":{"rendered":"

\u6807\u7b7e\u5e73\u6ed1\u662f\u673a\u5668\u5b66\u4e60\u548c\u6df1\u5ea6\u5b66\u4e60\u6a21\u578b\u4e2d\u5e38\u7528\u7684\u6b63\u5219\u5316\u6280\u672f\u3002\u5b83\u6d89\u53ca\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u5411\u76ee\u6807\u6807\u7b7e\u6dfb\u52a0\u5c11\u91cf\u7684\u4e0d\u786e\u5b9a\u6027\uff0c\u8fd9\u6709\u52a9\u4e8e\u9632\u6b62\u8fc7\u5ea6\u62df\u5408\u5e76\u63d0\u9ad8\u6a21\u578b\u7684\u6cdb\u5316\u80fd\u529b\u3002\u901a\u8fc7\u5f15\u5165\u66f4\u73b0\u5b9e\u7684\u6807\u7b7e\u5206\u5e03\u5f62\u5f0f\uff0c\u6807\u7b7e\u5e73\u6ed1\u53ef\u786e\u4fdd\u6a21\u578b\u51cf\u5c11\u5bf9\u5355\u4e2a\u6807\u7b7e\u786e\u5b9a\u6027\u7684\u4f9d\u8d56\uff0c\u4ece\u800c\u63d0\u9ad8\u672a\u89c1\u6570\u636e\u7684\u6027\u80fd\u3002<\/p>\n

\u6807\u7b7e\u5e73\u6ed1\u7684\u8d77\u6e90\u5386\u53f2\u53ca\u5176\u9996\u6b21\u63d0\u53ca<\/h2>\n

\u6807\u7b7e\u5e73\u6ed1\u9996\u6b21\u5728 Christian Szegedy \u7b49\u4eba\u4e8e 2016 \u5e74\u53d1\u8868\u7684\u9898\u4e3a\u201c\u91cd\u65b0\u601d\u8003\u8ba1\u7b97\u673a\u89c6\u89c9\u7684\u521d\u59cb\u67b6\u6784\u201d\u7684\u7814\u7a76\u8bba\u6587\u4e2d\u88ab\u5f15\u5165\u3002\u4f5c\u8005\u63d0\u51fa\u6807\u7b7e\u5e73\u6ed1\u4f5c\u4e3a\u4e00\u79cd\u89c4\u8303\u6df1\u5ea6\u5377\u79ef\u795e\u7ecf\u7f51\u7edc (CNN) \u5e76\u51cf\u8f7b\u8fc7\u5ea6\u62df\u5408\u7684\u4e0d\u5229\u5f71\u54cd\uff0c\u7279\u522b\u662f\u5728\u5927\u89c4\u6a21\u56fe\u50cf\u5206\u7c7b\u4efb\u52a1\u7684\u80cc\u666f\u4e0b\u3002<\/p>\n

\u6709\u5173\u6807\u7b7e\u5e73\u6ed1\u7684\u8be6\u7ec6\u4fe1\u606f\u3002\u6269\u5c55\u4e3b\u9898\u6807\u7b7e\u5e73\u6ed1\u3002<\/h2>\n

\u5728\u4f20\u7edf\u7684\u76d1\u7763\u5b66\u4e60\u4e2d\uff0c\u6a21\u578b\u7ecf\u8fc7\u8bad\u7ec3\u4ee5\u7edd\u5bf9\u786e\u5b9a\u5730\u8fdb\u884c\u9884\u6d4b\uff0c\u65e8\u5728\u6700\u5c0f\u5316\u9884\u6d4b\u6807\u7b7e\u548c\u771f\u5b9e\u6807\u7b7e\u4e4b\u95f4\u7684\u4ea4\u53c9\u71b5\u635f\u5931\u3002\u7136\u800c\uff0c\u8fd9\u79cd\u65b9\u6cd5\u53ef\u80fd\u4f1a\u5bfc\u81f4\u8fc7\u5ea6\u81ea\u4fe1\u7684\u9884\u6d4b\uff0c\u5373\u6a21\u578b\u5bf9\u4e0d\u6b63\u786e\u7684\u9884\u6d4b\u8fc7\u4e8e\u81ea\u4fe1\uff0c\u6700\u7ec8\u963b\u788d\u5176\u5bf9\u672a\u89c1\u6570\u636e\u7684\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n

\u6807\u7b7e\u5e73\u6ed1\u901a\u8fc7\u5728\u8bad\u7ec3\u671f\u95f4\u5f15\u5165\u4e00\u79cd\u8f6f\u6807\u7b7e\u5f62\u5f0f\u6765\u89e3\u51b3\u8fd9\u4e2a\u95ee\u9898\u3002\u6807\u7b7e\u5e73\u6ed1\u4e0d\u662f\u5206\u914d\u4e00\u4e2a one-hot \u7f16\u7801\u5411\u91cf\uff081 \u4ee3\u8868\u771f\u5b9e\u6807\u7b7e\uff0c0 \u4ee3\u8868\u5176\u4ed6\u6807\u7b7e\uff09\u4f5c\u4e3a\u76ee\u6807\uff0c\u800c\u662f\u5c06\u6982\u7387\u8d28\u91cf\u5206\u5e03\u5728\u6240\u6709\u7c7b\u522b\u4e4b\u95f4\u3002\u4e3a\u771f\u5b9e\u6807\u7b7e\u5206\u914d\u7565\u5c0f\u4e8e 1 \u7684\u6982\u7387\uff0c\u5e76\u5c06\u5269\u4f59\u6982\u7387\u5206\u914d\u7ed9\u5176\u4ed6\u7c7b\u522b\u3002\u8fd9\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u5f15\u5165\u4e86\u4e0d\u786e\u5b9a\u6027\uff0c\u4f7f\u6a21\u578b\u4e0d\u6613\u8fc7\u5ea6\u62df\u5408\u5e76\u4e14\u66f4\u52a0\u7a33\u5065\u3002<\/p>\n

Label\u5e73\u6ed1\u7684\u5185\u90e8\u7ed3\u6784\u3002\u6807\u7b7e\u5e73\u6ed1\u7684\u5de5\u4f5c\u539f\u7406\u3002<\/h2>\n

\u6807\u7b7e\u5e73\u6ed1\u7684\u5185\u90e8\u5de5\u4f5c\u53ef\u4ee5\u6982\u62ec\u4e3a\u4ee5\u4e0b\u51e0\u4e2a\u6b65\u9aa4\uff1a<\/p>\n

    \n
  1. \n

    \u4e00\u70ed\u7f16\u7801\uff1a<\/strong> \u5728\u4f20\u7edf\u7684\u76d1\u7763\u5b66\u4e60\u4e2d\uff0c\u6bcf\u4e2a\u6837\u672c\u7684\u76ee\u6807\u6807\u7b7e\u8868\u793a\u4e3a\u4e00\u4e2a one-hot \u7f16\u7801\u5411\u91cf\uff0c\u5176\u4e2d\u771f\u5b9e\u7c7b\u522b\u7684\u503c\u4e3a 1\uff0c\u6240\u6709\u5176\u4ed6\u7c7b\u522b\u7684\u503c\u4e3a 0\u3002<\/p>\n<\/li>\n

  2. \n

    \u8f6f\u5316\u6807\u7b7e\uff1a<\/strong> \u6807\u7b7e\u5e73\u6ed1\u901a\u8fc7\u5728\u6240\u6709\u7c7b\u4e4b\u95f4\u5206\u914d\u6982\u7387\u8d28\u91cf\u6765\u4fee\u6539\u72ec\u70ed\u7f16\u7801\u7684\u76ee\u6807\u6807\u7b7e\u3002\u5b83\u6ca1\u6709\u4e3a\u771f\u5b9e\u7c7b\u5206\u914d\u503c 1\uff0c\u800c\u662f\u5206\u914d\u503c (1 \u2013 \u03b5)\uff0c\u5176\u4e2d \u03b5 \u662f\u4e00\u4e2a\u5c0f\u7684\u6b63\u5e38\u6570\u3002<\/p>\n<\/li>\n

  3. \n

    \u5206\u5e03\u4e0d\u786e\u5b9a\u6027\uff1a<\/strong> \u5269\u4f59\u7684\u6982\u7387 \u03b5 \u88ab\u5212\u5206\u5230\u5176\u4ed6\u7c7b\u522b\u4e2d\uff0c\u4f7f\u6a21\u578b\u8003\u8651\u8fd9\u4e9b\u7c7b\u522b\u662f\u6b63\u786e\u7c7b\u522b\u7684\u53ef\u80fd\u6027\u3002\u8fd9\u5f15\u5165\u4e86\u4e00\u5b9a\u7a0b\u5ea6\u7684\u4e0d\u786e\u5b9a\u6027\uff0c\u5bfc\u81f4\u6a21\u578b\u5bf9\u5176\u9884\u6d4b\u7684\u786e\u5b9a\u6027\u964d\u4f4e\u3002<\/p>\n<\/li>\n

  4. \n

    \u635f\u8017\u8ba1\u7b97\uff1a<\/strong> \u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u6a21\u578b\u4f18\u5316\u9884\u6d4b\u6982\u7387\u548c\u8f6f\u5316\u76ee\u6807\u6807\u7b7e\u4e4b\u95f4\u7684\u4ea4\u53c9\u71b5\u635f\u5931\u3002\u6807\u7b7e\u5e73\u6ed1\u635f\u5931\u4f1a\u60e9\u7f5a\u8fc7\u5ea6\u81ea\u4fe1\u7684\u9884\u6d4b\u5e76\u4fc3\u8fdb\u66f4\u52a0\u6821\u51c6\u7684\u9884\u6d4b\u3002<\/p>\n<\/li>\n<\/ol>\n

    \u6807\u7b7e\u5e73\u6ed1\u7684\u5173\u952e\u7279\u5f81\u5206\u6790\u3002<\/h2>\n

    \u6807\u7b7e\u5e73\u6ed1\u7684\u4e3b\u8981\u529f\u80fd\u5305\u62ec\uff1a<\/p>\n

      \n
    1. \n

      \u6b63\u5219\u5316\uff1a<\/strong> \u6807\u7b7e\u5e73\u6ed1\u4f5c\u4e3a\u4e00\u79cd\u6b63\u5219\u5316\u6280\u672f\uff0c\u53ef\u4ee5\u9632\u6b62\u8fc7\u5ea6\u62df\u5408\u5e76\u63d0\u9ad8\u6a21\u578b\u6cdb\u5316\u80fd\u529b\u3002<\/p>\n<\/li>\n

    2. \n

      \u6821\u51c6\u9884\u6d4b\uff1a<\/strong> \u901a\u8fc7\u5728\u76ee\u6807\u6807\u7b7e\u4e2d\u5f15\u5165\u4e0d\u786e\u5b9a\u6027\uff0c\u6807\u7b7e\u5e73\u6ed1\u53ef\u4ee5\u9f13\u52b1\u6a21\u578b\u4ea7\u751f\u66f4\u52a0\u6821\u51c6\u4e14\u4e0d\u592a\u81ea\u4fe1\u7684\u9884\u6d4b\u3002<\/p>\n<\/li>\n

    3. \n

      \u63d0\u9ad8\u7a33\u5065\u6027\uff1a<\/strong> \u6807\u7b7e\u5e73\u6ed1\u6709\u52a9\u4e8e\u6a21\u578b\u4e13\u6ce8\u4e8e\u5b66\u4e60\u6570\u636e\u4e2d\u6709\u610f\u4e49\u7684\u6a21\u5f0f\uff0c\u800c\u4e0d\u662f\u8bb0\u4f4f\u7279\u5b9a\u7684\u8bad\u7ec3\u6837\u672c\uff0c\u4ece\u800c\u63d0\u9ad8\u9c81\u68d2\u6027\u3002<\/p>\n<\/li>\n

    4. \n

      \u5904\u7406\u6709\u566a\u97f3\u7684\u6807\u7b7e\uff1a<\/strong> \u6807\u7b7e\u5e73\u6ed1\u53ef\u4ee5\u6bd4\u4f20\u7edf\u7684\u5355\u70ed\u7f16\u7801\u76ee\u6807\u66f4\u6709\u6548\u5730\u5904\u7406\u566a\u58f0\u6216\u4e0d\u6b63\u786e\u7684\u6807\u7b7e\u3002<\/p>\n<\/li>\n<\/ol>\n

      \u6807\u7b7e\u5e73\u6ed1\u7684\u7c7b\u578b<\/h2>\n

      \u6807\u7b7e\u5e73\u6ed1\u6709\u4e24\u79cd\u5e38\u89c1\u7c7b\u578b\uff1a<\/p>\n

        \n
      1. \n

        \u56fa\u5b9a\u6807\u7b7e\u5e73\u6ed1\uff1a<\/strong> \u5728\u8fd9\u79cd\u65b9\u6cd5\u4e2d\uff0c\u03b5\uff08\u7528\u4e8e\u8f6f\u5316\u771f\u5b9e\u6807\u7b7e\u7684\u5e38\u6570\uff09\u7684\u503c\u5728\u6574\u4e2a\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u662f\u56fa\u5b9a\u7684\u3002\u5bf9\u4e8e\u6570\u636e\u96c6\u4e2d\u7684\u6240\u6709\u6837\u672c\uff0c\u5b83\u4fdd\u6301\u4e0d\u53d8\u3002<\/p>\n<\/li>\n

      2. \n

        \u9000\u706b\u6807\u7b7e\u5e73\u6ed1\uff1a<\/strong> \u4e0e\u56fa\u5b9a\u6807\u7b7e\u5e73\u6ed1\u4e0d\u540c\uff0c\u03b5 \u7684\u503c\u5728\u8bad\u7ec3\u671f\u95f4\u9000\u706b\u6216\u8870\u51cf\u3002\u5b83\u4ece\u8f83\u9ad8\u7684\u503c\u5f00\u59cb\uff0c\u5e76\u968f\u7740\u8bad\u7ec3\u7684\u8fdb\u884c\u9010\u6e10\u51cf\u5c0f\u3002\u8fd9\u4f7f\u5f97\u6a21\u578b\u80fd\u591f\u4ece\u8f83\u9ad8\u6c34\u5e73\u7684\u4e0d\u786e\u5b9a\u6027\u5f00\u59cb\uff0c\u5e76\u968f\u7740\u65f6\u95f4\u7684\u63a8\u79fb\u800c\u964d\u4f4e\uff0c\u4ece\u800c\u6709\u6548\u5730\u5fae\u8c03\u9884\u6d4b\u7684\u6821\u51c6\u3002<\/p>\n<\/li>\n<\/ol>\n

        \u8fd9\u4e9b\u7c7b\u578b\u4e4b\u95f4\u7684\u9009\u62e9\u53d6\u51b3\u4e8e\u7279\u5b9a\u4efb\u52a1\u548c\u6570\u636e\u96c6\u7279\u5f81\u3002\u56fa\u5b9a\u6807\u7b7e\u5e73\u6ed1\u66f4\u5bb9\u6613\u5b9e\u73b0\uff0c\u800c\u9000\u706b\u6807\u7b7e\u5e73\u6ed1\u53ef\u80fd\u9700\u8981\u8c03\u6574\u8d85\u53c2\u6570\u624d\u80fd\u5b9e\u73b0\u6700\u4f73\u6027\u80fd\u3002<\/p>\n

        \u4e0b\u9762\u662f\u4e24\u79cd\u6807\u7b7e\u5e73\u6ed1\u7684\u6bd4\u8f83\uff1a<\/p>\n\n\n\n\n\n\n\n\n
        \u65b9\u9762<\/th>\n\u56fa\u5b9a\u6807\u7b7e\u5e73\u6ed1<\/th>\n\u9000\u706b\u6807\u7b7e\u5e73\u6ed1<\/th>\n<\/tr>\n<\/thead>\n
        \u03b5\u503c<\/td>\n\u59cb\u7ec8\u5982\u4e00<\/td>\n\u9000\u706b\u6216\u8150\u70c2<\/td>\n<\/tr>\n
        \u590d\u6742<\/td>\n\u5b9e\u65bd\u8d77\u6765\u66f4\u7b80\u5355<\/td>\n\u53ef\u80fd\u9700\u8981\u8d85\u53c2\u6570\u8c03\u6574<\/td>\n<\/tr>\n
        \u6821\u51c6<\/td>\n\u5fae\u8c03\u8f83\u5c11<\/td>\n\u968f\u7740\u65f6\u95f4\u7684\u63a8\u79fb\u9010\u6e10\u6539\u5584<\/td>\n<\/tr>\n
        \u8868\u73b0<\/td>\n\u6027\u80fd\u7a33\u5b9a<\/td>\n\u53d6\u5f97\u66f4\u597d\u7ed3\u679c\u7684\u6f5c\u529b<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

        \u6807\u7b7e\u5e73\u6ed1\u7684\u4f7f\u7528\u65b9\u6cd5\u3001\u95ee\u9898\u4ee5\u53ca\u4e0e\u4f7f\u7528\u76f8\u5173\u7684\u89e3\u51b3\u65b9\u6848\u3002<\/h2>\n

        \u4f7f\u7528\u6807\u7b7e\u5e73\u6ed1<\/h3>\n

        \u6807\u7b7e\u5e73\u6ed1\u53ef\u4ee5\u8f7b\u677e\u5730\u878d\u5165\u5404\u79cd\u673a\u5668\u5b66\u4e60\u6a21\u578b\u7684\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u5305\u62ec\u795e\u7ecf\u7f51\u7edc\u548c\u6df1\u5ea6\u5b66\u4e60\u67b6\u6784\u3002\u5b83\u6d89\u53ca\u5728\u6bcf\u6b21\u8bad\u7ec3\u8fed\u4ee3\u671f\u95f4\u8ba1\u7b97\u635f\u5931\u4e4b\u524d\u4fee\u6539\u76ee\u6807\u6807\u7b7e\u3002<\/p>\n

        \u5b9e\u65bd\u6b65\u9aa4\u5982\u4e0b\uff1a<\/p>\n

          \n
        1. \u4f7f\u7528 one-hot \u7f16\u7801\u76ee\u6807\u6807\u7b7e\u51c6\u5907\u6570\u636e\u96c6\u3002<\/li>\n
        2. \u6839\u636e\u5b9e\u9a8c\u6216\u9886\u57df\u4e13\u4e1a\u77e5\u8bc6\u5b9a\u4e49\u6807\u7b7e\u5e73\u6ed1\u503c \u03b5\u3002<\/li>\n
        3. \u5982\u524d\u6240\u8ff0\uff0c\u901a\u8fc7\u5206\u5e03\u6982\u7387\u8d28\u91cf\uff0c\u5c06\u72ec\u70ed\u7f16\u7801\u6807\u7b7e\u8f6c\u6362\u4e3a\u8f6f\u5316\u6807\u7b7e\u3002<\/li>\n
        4. \u4f7f\u7528\u8f6f\u5316\u6807\u7b7e\u8bad\u7ec3\u6a21\u578b\uff0c\u5e76\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u4f18\u5316\u4ea4\u53c9\u71b5\u635f\u5931\u3002<\/li>\n<\/ol>\n

          \u95ee\u9898\u4e0e\u89e3\u51b3\u65b9\u6848<\/h3>\n

          \u867d\u7136\u6807\u7b7e\u5e73\u6ed1\u63d0\u4f9b\u4e86\u591a\u79cd\u597d\u5904\uff0c\u4f46\u5b83\u4e5f\u53ef\u80fd\u5e26\u6765\u67d0\u4e9b\u6311\u6218\uff1a<\/p>\n

            \n
          1. \n

            \u5bf9\u51c6\u786e\u6027\u7684\u5f71\u54cd\uff1a<\/strong> \u5728\u67d0\u4e9b\u60c5\u51b5\u4e0b\uff0c\u7531\u4e8e\u5f15\u5165\u4e86\u4e0d\u786e\u5b9a\u6027\uff0c\u6807\u7b7e\u5e73\u6ed1\u53ef\u80fd\u4f1a\u7a0d\u5fae\u964d\u4f4e\u6a21\u578b\u5728\u8bad\u7ec3\u96c6\u4e0a\u7684\u51c6\u786e\u6027\u3002\u7136\u800c\uff0c\u5b83\u901a\u5e38\u4f1a\u63d0\u9ad8\u6d4b\u8bd5\u96c6\u6216\u672a\u89c1\u8fc7\u7684\u6570\u636e\u7684\u6027\u80fd\uff0c\u8fd9\u662f\u6807\u7b7e\u5e73\u6ed1\u7684\u4e3b\u8981\u76ee\u6807\u3002<\/p>\n<\/li>\n

          2. \n

            \u8d85\u53c2\u6570\u8c03\u4f18\uff1a<\/strong> \u9009\u62e9\u5408\u9002\u7684 \u03b5 \u503c\u5bf9\u4e8e\u6709\u6548\u7684\u6807\u7b7e\u5e73\u6ed1\u81f3\u5173\u91cd\u8981\u3002\u8fc7\u9ad8\u6216\u8fc7\u4f4e\u7684\u503c\u53ef\u80fd\u4f1a\u5bf9\u6a21\u578b\u7684\u6027\u80fd\u4ea7\u751f\u8d1f\u9762\u5f71\u54cd\u3002\u8d85\u53c2\u6570\u8c03\u6574\u6280\u672f\uff08\u4f8b\u5982\u7f51\u683c\u641c\u7d22\u6216\u968f\u673a\u641c\u7d22\uff09\u53ef\u7528\u4e8e\u627e\u5230\u6700\u4f73 \u03b5 \u503c\u3002<\/p>\n<\/li>\n

          3. \n

            \u635f\u5931\u51fd\u6570\u4fee\u6539\uff1a<\/strong> \u5b9e\u73b0\u6807\u7b7e\u5e73\u6ed1\u9700\u8981\u4fee\u6539\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u7684\u635f\u5931\u51fd\u6570\u3002\u6b64\u4fee\u6539\u53ef\u80fd\u4f1a\u4f7f\u8bad\u7ec3\u7ba1\u9053\u590d\u6742\u5316\uff0c\u5e76\u4e14\u9700\u8981\u5bf9\u73b0\u6709\u4ee3\u7801\u5e93\u8fdb\u884c\u8c03\u6574\u3002<\/p>\n<\/li>\n<\/ol>\n

            \u4e3a\u4e86\u7f13\u89e3\u8fd9\u4e9b\u95ee\u9898\uff0c\u7814\u7a76\u4eba\u5458\u548c\u4ece\u4e1a\u8005\u53ef\u4ee5\u5c1d\u8bd5\u4e0d\u540c\u7684 \u03b5 \u503c\uff0c\u76d1\u63a7\u6a21\u578b\u5728\u9a8c\u8bc1\u6570\u636e\u4e0a\u7684\u6027\u80fd\uff0c\u5e76\u76f8\u5e94\u5730\u5fae\u8c03\u8d85\u53c2\u6570\u3002\u6b64\u5916\uff0c\u5f7b\u5e95\u7684\u6d4b\u8bd5\u548c\u5b9e\u9a8c\u5bf9\u4e8e\u8bc4\u4f30\u6807\u7b7e\u5e73\u6ed1\u5bf9\u7279\u5b9a\u4efb\u52a1\u548c\u6570\u636e\u96c6\u7684\u5f71\u54cd\u81f3\u5173\u91cd\u8981\u3002<\/p>\n

            \u4ee5\u8868\u683c\u548c\u5217\u8868\u7684\u5f62\u5f0f\u5217\u51fa\u4e3b\u8981\u7279\u5f81\u4ee5\u53ca\u4e0e\u7c7b\u4f3c\u672f\u8bed\u7684\u5176\u4ed6\u6bd4\u8f83\u3002<\/h2>\n

            \u4e0b\u9762\u662f\u6807\u7b7e\u5e73\u6ed1\u4e0e\u5176\u4ed6\u76f8\u5173\u6b63\u5219\u5316\u6280\u672f\u7684\u6bd4\u8f83\uff1a<\/p>\n\n\n\n\n\n\n\n\n
            \u6b63\u5219\u5316\u6280\u672f<\/th>\n\u7279\u5f81<\/th>\n<\/tr>\n<\/thead>\n
            L1 \u548c L2 \u6b63\u5219\u5316<\/td>\n\u60e9\u7f5a\u6a21\u578b\u4e2d\u7684\u5927\u6743\u91cd\u4ee5\u9632\u6b62\u8fc7\u5ea6\u62df\u5408\u3002<\/td>\n<\/tr>\n
            \u8f8d\u5b66<\/td>\n\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u968f\u673a\u505c\u7528\u795e\u7ecf\u5143\u4ee5\u9632\u6b62\u8fc7\u5ea6\u62df\u5408\u3002<\/td>\n<\/tr>\n
            \u6570\u636e\u589e\u5f3a<\/td>\n\u5f15\u5165\u8bad\u7ec3\u6570\u636e\u7684\u53d8\u4f53\u4ee5\u589e\u52a0\u6570\u636e\u96c6\u5927\u5c0f\u3002<\/td>\n<\/tr>\n
            \u6807\u7b7e\u5e73\u6ed1<\/td>\n\u8f6f\u5316\u76ee\u6807\u6807\u7b7e\u4ee5\u9f13\u52b1\u6821\u51c6\u9884\u6d4b\u3002<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n

            \u867d\u7136\u6240\u6709\u8fd9\u4e9b\u6280\u672f\u90fd\u65e8\u5728\u63d0\u9ad8\u6a21\u578b\u6cdb\u5316\u80fd\u529b\uff0c\u4f46\u6807\u7b7e\u5e73\u6ed1\u56e0\u5176\u91cd\u70b9\u5173\u6ce8\u5728\u76ee\u6807\u6807\u7b7e\u4e2d\u5f15\u5165\u4e0d\u786e\u5b9a\u6027\u800c\u8131\u9896\u800c\u51fa\u3002\u5b83\u6709\u52a9\u4e8e\u6a21\u578b\u505a\u51fa\u66f4\u52a0\u81ea\u4fe1\u800c\u8c28\u614e\u7684\u9884\u6d4b\uff0c\u4ece\u800c\u5728\u672a\u89c1\u8fc7\u7684\u6570\u636e\u4e0a\u83b7\u5f97\u66f4\u597d\u7684\u6027\u80fd\u3002<\/p>\n

            \u4e0e\u6807\u7b7e\u5e73\u6ed1\u76f8\u5173\u7684\u672a\u6765\u89c2\u70b9\u548c\u6280\u672f\u3002<\/h2>\n

            \u6df1\u5ea6\u5b66\u4e60\u548c\u673a\u5668\u5b66\u4e60\u9886\u57df\uff0c\u5305\u62ec\u6807\u7b7e\u5e73\u6ed1\u7b49\u6b63\u5219\u5316\u6280\u672f\uff0c\u6b63\u5728\u4e0d\u65ad\u53d1\u5c55\u3002\u7814\u7a76\u4eba\u5458\u6b63\u5728\u63a2\u7d22\u66f4\u5148\u8fdb\u7684\u6b63\u5219\u5316\u65b9\u6cd5\u53ca\u5176\u7ec4\u5408\uff0c\u4ee5\u8fdb\u4e00\u6b65\u63d0\u9ad8\u6a21\u578b\u6027\u80fd\u548c\u6cdb\u5316\u80fd\u529b\u3002\u6807\u7b7e\u5e73\u6ed1\u53ca\u76f8\u5173\u9886\u57df\u672a\u6765\u7814\u7a76\u7684\u4e00\u4e9b\u6f5c\u5728\u65b9\u5411\u5305\u62ec\uff1a<\/p>\n

              \n
            1. \n

              \u81ea\u9002\u5e94\u6807\u7b7e\u5e73\u6ed1\uff1a<\/strong> \u7814\u7a76\u6839\u636e\u6a21\u578b\u9884\u6d4b\u7684\u7f6e\u4fe1\u5ea6\u52a8\u6001\u8c03\u6574 \u03b5 \u503c\u7684\u6280\u672f\u3002\u8fd9\u53ef\u80fd\u4f1a\u5bfc\u81f4\u8bad\u7ec3\u671f\u95f4\u66f4\u5177\u9002\u5e94\u6027\u7684\u4e0d\u786e\u5b9a\u6027\u6c34\u5e73\u3002<\/p>\n<\/li>\n

            2. \n

              \u7279\u5b9a\u9886\u57df\u7684\u6807\u7b7e\u5e73\u6ed1\uff1a<\/strong> \u9488\u5bf9\u7279\u5b9a\u9886\u57df\u6216\u4efb\u52a1\u5b9a\u5236\u6807\u7b7e\u5e73\u6ed1\u6280\u672f\uff0c\u4ee5\u8fdb\u4e00\u6b65\u63d0\u9ad8\u5176\u6709\u6548\u6027\u3002<\/p>\n<\/li>\n

            3. \n

              \u4e0e\u5176\u4ed6\u6b63\u5219\u5316\u6280\u672f\u7684\u76f8\u4e92\u4f5c\u7528\uff1a<\/strong> \u63a2\u7d22\u6807\u7b7e\u5e73\u6ed1\u548c\u5176\u4ed6\u6b63\u5219\u5316\u65b9\u6cd5\u4e4b\u95f4\u7684\u534f\u540c\u4f5c\u7528\uff0c\u4ee5\u5728\u590d\u6742\u6a21\u578b\u4e2d\u5b9e\u73b0\u66f4\u597d\u7684\u6cdb\u5316\u3002<\/p>\n<\/li>\n

            4. \n

              \u5f3a\u5316\u5b66\u4e60\u4e2d\u7684\u6807\u7b7e\u5e73\u6ed1\uff1a<\/strong> \u5c06\u6807\u7b7e\u5e73\u6ed1\u6280\u672f\u6269\u5c55\u5230\u5f3a\u5316\u5b66\u4e60\u9886\u57df\uff0c\u5176\u4e2d\u5956\u52b1\u7684\u4e0d\u786e\u5b9a\u6027\u53ef\u4ee5\u53d1\u6325\u81f3\u5173\u91cd\u8981\u7684\u4f5c\u7528\u3002<\/p>\n<\/li>\n<\/ol>\n

              \u5982\u4f55\u4f7f\u7528\u4ee3\u7406\u670d\u52a1\u5668\u6216\u5982\u4f55\u5c06\u4ee3\u7406\u670d\u52a1\u5668\u4e0e\u6807\u7b7e\u5e73\u6ed1\u5173\u8054\u3002<\/h2>\n

              \u4ee3\u7406\u670d\u52a1\u5668\u548c\u6807\u7b7e\u5e73\u6ed1\u5e76\u4e0d\u76f4\u63a5\u76f8\u5173\uff0c\u56e0\u4e3a\u5b83\u4eec\u5728\u6280\u672f\u9886\u57df\u670d\u52a1\u4e8e\u4e0d\u540c\u7684\u76ee\u7684\u3002\u7136\u800c\uff0c\u4ee3\u7406\u670d\u52a1\u5668\u53ef\u4ee5\u4e0e\u673a\u5668\u5b66\u4e60\u6a21\u578b\u7ed3\u5408\u4f7f\u7528\uff0c\u4ee5\u5404\u79cd\u65b9\u5f0f\u5b9e\u73b0\u6807\u7b7e\u5e73\u6ed1\uff1a<\/p>\n

                \n
              1. \n

                \u6570\u636e\u91c7\u96c6\uff1a<\/strong> \u4ee3\u7406\u670d\u52a1\u5668\u53ef\u7528\u4e8e\u6536\u96c6\u6765\u81ea\u4e0d\u540c\u5730\u7406\u4f4d\u7f6e\u7684\u4e0d\u540c\u6570\u636e\u96c6\uff0c\u786e\u4fdd\u673a\u5668\u5b66\u4e60\u6a21\u578b\u7684\u8bad\u7ec3\u6570\u636e\u80fd\u591f\u4ee3\u8868\u4e0d\u540c\u7684\u7528\u6237\u7fa4\u4f53\u3002<\/p>\n<\/li>\n

              2. \n

                \u533f\u540d\u548c\u9690\u79c1\uff1a<\/strong> \u4ee3\u7406\u670d\u52a1\u5668\u53ef\u7528\u4e8e\u5728\u6570\u636e\u6536\u96c6\u8fc7\u7a0b\u4e2d\u5bf9\u7528\u6237\u6570\u636e\u8fdb\u884c\u533f\u540d\u5316\uff0c\u4ece\u800c\u89e3\u51b3\u5728\u654f\u611f\u4fe1\u606f\u4e0a\u8bad\u7ec3\u6a21\u578b\u65f6\u7684\u9690\u79c1\u95ee\u9898\u3002<\/p>\n<\/li>\n

              3. \n

                \u6a21\u578b\u670d\u52a1\u7684\u8d1f\u8f7d\u5e73\u8861\uff1a<\/strong> \u5728\u90e8\u7f72\u9636\u6bb5\uff0c\u4ee3\u7406\u670d\u52a1\u5668\u53ef\u7528\u4e8e\u5728\u673a\u5668\u5b66\u4e60\u6a21\u578b\u7684\u591a\u4e2a\u5b9e\u4f8b\u4e4b\u95f4\u6709\u6548\u5730\u8fdb\u884c\u8d1f\u8f7d\u5e73\u8861\u548c\u5206\u53d1\u6a21\u578b\u63a8\u7406\u8bf7\u6c42\u3002<\/p>\n<\/li>\n

              4. \n

                \u7f13\u5b58\u6a21\u578b\u9884\u6d4b\uff1a<\/strong> \u4ee3\u7406\u670d\u52a1\u5668\u53ef\u4ee5\u7f13\u5b58\u673a\u5668\u5b66\u4e60\u6a21\u578b\u505a\u51fa\u7684\u9884\u6d4b\uff0c\u4ece\u800c\u51cf\u5c11\u91cd\u590d\u67e5\u8be2\u7684\u54cd\u5e94\u65f6\u95f4\u548c\u670d\u52a1\u5668\u8d1f\u8f7d\u3002<\/p>\n<\/li>\n<\/ol>\n

                \u867d\u7136\u4ee3\u7406\u670d\u52a1\u5668\u548c\u6807\u7b7e\u5e73\u6ed1\u72ec\u7acb\u8fd0\u884c\uff0c\u4f46\u524d\u8005\u53ef\u4ee5\u5728\u786e\u4fdd\u7a33\u5065\u7684\u6570\u636e\u6536\u96c6\u548c\u6709\u6548\u90e8\u7f72\u4f7f\u7528\u6807\u7b7e\u5e73\u6ed1\u6280\u672f\u8bad\u7ec3\u7684\u673a\u5668\u5b66\u4e60\u6a21\u578b\u65b9\u9762\u53d1\u6325\u652f\u6301\u4f5c\u7528\u3002<\/p>\n

                \u76f8\u5173\u94fe\u63a5<\/h2>\n

                \u6709\u5173\u6807\u7b7e\u5e73\u6ed1\u53ca\u5176\u5728\u6df1\u5ea6\u5b66\u4e60\u4e2d\u7684\u5e94\u7528\u7684\u66f4\u591a\u4fe1\u606f\uff0c\u8bf7\u8003\u8651\u63a2\u7d22\u4ee5\u4e0b\u8d44\u6e90\uff1a<\/p>\n

                  \n
                1. \u91cd\u65b0\u601d\u8003\u8ba1\u7b97\u673a\u89c6\u89c9\u7684 Inception \u67b6\u6784<\/a> \u2013 \u4ecb\u7ecd\u6807\u7b7e\u5e73\u6ed1\u7684\u539f\u59cb\u7814\u7a76\u8bba\u6587\u3002<\/li>\n
                2. \u6807\u7b7e\u5e73\u6ed1\u7684\u7b80\u5355\u4ecb\u7ecd<\/a> \u2013 \u9488\u5bf9\u521d\u5b66\u8005\u7684\u6807\u7b7e\u5e73\u6ed1\u8be6\u7ec6\u6559\u7a0b\u3002<\/li>\n
                3. \u4e86\u89e3\u6807\u7b7e\u5e73\u6ed1<\/a> \u2013 \u6807\u7b7e\u5e73\u6ed1\u53ca\u5176\u5bf9\u6a21\u578b\u8bad\u7ec3\u7684\u5f71\u54cd\u7684\u5168\u9762\u89e3\u91ca\u3002<\/li>\n<\/ol>","protected":false},"featured_media":468749,"menu_order":0,"template":"","meta":{"_acf_changed":false,"content-type":"","inline_featured_image":false,"footnotes":""},"class_list":["post-477793","wiki","type-wiki","status-publish","has-post-thumbnail","hentry"],"acf":{"faq_title":"Frequently Asked Questions about Label Smoothing<\/mark>","faq_items":[{"question":"What is Label Smoothing?","answer":"

                  Label smoothing is a regularization technique used in machine learning and deep learning models. It involves adding a small amount of uncertainty to the target labels during training to prevent overfitting and improve model generalization.<\/p>"},{"question":"How was Label Smoothing introduced?","answer":"

                  Label smoothing was first introduced in the research paper \"Rethinking the Inception Architecture for Computer Vision\" by Christian Szegedy et al. in 2016. The authors proposed it as a regularization method for large-scale image classification tasks.<\/p>"},{"question":"How does Label Smoothing work?","answer":"

                  Label smoothing modifies the traditional one-hot encoded target labels by distributing the probability mass among all classes. The true label is assigned a value slightly less than one, and the remaining probabilities are divided among other classes, introducing a sense of uncertainty during training.<\/p>"},{"question":"What are the types of Label Smoothing?","answer":"

                  There are two common types of label smoothing: fixed label smoothing and annealing label smoothing. Fixed label smoothing uses a constant value for uncertainty throughout training, while annealing label smoothing gradually decreases the uncertainty over time.<\/p>"},{"question":"How can I use Label Smoothing?","answer":"

                  To use label smoothing, modify the target labels before computing the loss during training. Prepare the dataset with one-hot encoded labels, choose a value for uncertainty (\u03b5), and convert the labels into softened labels with the probability distribution.<\/p>"},{"question":"What benefits does Label Smoothing offer?","answer":"

                  Label smoothing improves model robustness and calibration, making it less reliant on individual labels during prediction. It also handles noisy labels better and enhances generalization performance on unseen data.<\/p>"},{"question":"Are there any challenges with Label Smoothing?","answer":"

                  While label smoothing improves generalization, it might slightly reduce accuracy on the training set. Choosing an appropriate \u03b5 value requires experimentation, and implementation may need modification of the loss function.<\/p>"},{"question":"How can Proxy Servers be associated with Label Smoothing?","answer":"

                  Proxy servers are not directly related to label smoothing but can complement it. They can aid in diverse data collection, anonymizing user data, load balancing for model serving, and caching model predictions to optimize performance.<\/p>"}]},"_links":{"self":[{"href":"https:\/\/oneproxy.pro\/cn\/wp-json\/wp\/v2\/wiki\/477793","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/oneproxy.pro\/cn\/wp-json\/wp\/v2\/wiki"}],"about":[{"href":"https:\/\/oneproxy.pro\/cn\/wp-json\/wp\/v2\/types\/wiki"}],"version-history":[{"count":0,"href":"https:\/\/oneproxy.pro\/cn\/wp-json\/wp\/v2\/wiki\/477793\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/oneproxy.pro\/cn\/wp-json\/wp\/v2\/media\/468749"}],"wp:attachment":[{"href":"https:\/\/oneproxy.pro\/cn\/wp-json\/wp\/v2\/media?parent=477793"}],"curies":[{"name":"\u53ef\u6e7f\u6027\u7c89\u5242","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}