{"id":1143911,"date":"2025-07-03T18:16:22","date_gmt":"2025-07-04T01:16:22","guid":{"rendered":"https:\/\/www.microsoft.com\/en-us\/research\/?post_type=msr-blog-post&#038;p=1143911"},"modified":"2025-07-03T18:16:50","modified_gmt":"2025-07-04T01:16:50","slug":"ui-evol","status":"publish","type":"msr-blog-post","link":"https:\/\/www.microsoft.com\/en-us\/research\/articles\/ui-evol\/","title":{"rendered":"UI-Evol: \u8ba9\u667a\u80fd\u4f53\u201c\u77e5\u884c\u5408\u4e00\u201d"},"content":{"rendered":"\n<p>\u7f16\u8005\u6309\uff1a\u5f53 AI \u667a\u80fd\u4f53\u8d70\u51fa\u5b9e\u9a8c\u5ba4\uff0c\u5728\u771f\u5b9e\u4e16\u754c\u4e2d\u6267\u884c\u4efb\u52a1\u65f6\uff0c\u53ea\u6709\u77e5\u8bc6\u8fd8\u8fdc\u8fdc\u4e0d\u591f\u3002\u4f46\u5982\u4eca\u7684\u8bb8\u591a\u667a\u80fd\u4f53\u5373\u4f7f\u62e5\u6709\u6b63\u786e\u7684\u4fe1\u606f\uff0c\u53ef\u5728\u5b9e\u9645\u64cd\u4f5c\u4e2d\u4ecd\u9891\u9891\u5931\u8bef\uff0c\u4efb\u52a1\u5b8c\u6210\u7387\u4ee4\u4eba\u582a\u5fe7\u3002\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u5bf9\u6b64\u63d0\u51fa\u4e86\u4e00\u4e2a\u65b0\u65b9\u6cd5 UI-Evol\uff0c\u65e8\u5728\u5e2e\u52a9\u8ba1\u7b97\u673a\u4f7f\u7528\u667a\u80fd\u4f53\uff08Computer Use Agent\uff09\u8fdb\u5316\u77e5\u8bc6\uff0c\u4f7f\u667a\u80fd\u4f53\u4e0d\u4ec5\u201c\u77e5\u9053\u8be5\u600e\u4e48\u505a\u201d\uff0c\u8fd8\u8981\u771f\u6b63\u201c\u505a\u5f97\u5230\u201d\u3002\u5728\u8be5\u65b9\u6cd5\u7684\u52a0\u6301\u4e0b\uff0c\u667a\u80fd\u4f53\u5728\u5b9e\u9645\u6267\u884c\u80fd\u529b\u3001\u7a33\u5b9a\u6027\u548c\u53ef\u9760\u6027\u65b9\u9762\u5747\u5f97\u5230\u663e\u8457\u63d0\u5347\u3002\u76f8\u5173\u8bba\u6587\u88ab ICML 2025 Computer Use Workshop \u63a5\u6536\u3002<\/p>\n\n\n\n<hr class=\"wp-block-separator has-alpha-channel-opacity\"\/>\n\n\n\n<p>\u5728\u5f53\u524d\u8ba1\u7b97\u673a\u4f7f\u7528\u667a\u80fd\u4f53\uff08Computer Use Agent\uff09\u7684\u5f00\u53d1\u4e2d\uff0c\u5916\u90e8\u77e5\u8bc6\u626e\u6f14\u7740\u81f3\u5173\u91cd\u8981\u7684\u89d2\u8272\u3002\u7136\u800c\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u53d1\u73b0\u4e86\u4e00\u4e2a\u5173\u952e\u7684\u201c\u77e5\u8bc6-\u6267\u884c\u201d\u9e3f\u6c9f\uff0c\u5373\u6240\u68c0\u7d22\u5230\u7684\u77e5\u8bc6\u5f80\u5f80\u672a\u80fd\u6709\u6548\u5730\u8f6c\u5316\u4e3a\u5b9e\u9645\u7684\u4efb\u52a1\u6267\u884c\u3002\u8c03\u67e5\u5206\u6790\u663e\u793a\uff0c\u5373\u4f7f\u63d0\u4f9b\u7ed9\u667a\u80fd\u4f53\uff08agent\uff0990%\u7684\u6b63\u786e\u77e5\u8bc6\uff0c\u5176\u6267\u884c\u4efb\u52a1\u7684\u6210\u529f\u7387\u4e5f\u53ea\u670941%\u3002\u8fd9\u4e00\u5dee\u8ddd\u66b4\u9732\u51fa\u4e00\u4e2a\u6df1\u5c42\u95ee\u9898\uff1a\u5c3d\u7ba1\u667a\u80fd\u4f53\u62e5\u6709\u76f8\u5173\u77e5\u8bc6\uff0c\u4f46\u5728\u771f\u5b9e\u7684\u8ba1\u7b97\u673a\u4f7f\u7528\u73af\u5883\u4e2d\uff0c\u8fd9\u4e9b\u77e5\u8bc6\u5374\u96be\u4ee5\u88ab\u6709\u6548\u5e94\u7528\uff0c\u5bfc\u81f4\u4efb\u52a1\u5b8c\u6210\u7387\u4f4e\u4e0b\uff0c\u77e5\u8bc6\u4e0e\u884c\u52a8\u4e4b\u95f4\u5b58\u5728\u660e\u663e\u7684\u8131\u8282\u3002\u4e5f\u5c31\u662f\u8bf4\uff0c\u667a\u80fd\u4f53\u867d\u201c\u77e5\u5176\u7136\u201d\uff0c\u5374\u672a\u5fc5\u80fd\u201c\u884c\u5176\u662f\u201d\u3002<\/p>\n\n\n\n<p>\u901a\u8fc7\u5b9e\u9a8c\uff0c\u7814\u7a76\u5458\u4eec\u8fdb\u4e00\u6b65\u53d1\u73b0\uff0c\u73b0\u9636\u6bb5 Computer Use Agent \u6267\u884c\u4efb\u52a1\u7684\u6210\u529f\u7387\u5b58\u5728\u8f83\u5f3a\u7684\u968f\u673a\u6027\uff0c\u4e25\u91cd\u5f71\u54cd\u7528\u6237\u7684\u5b9e\u9645\u4f53\u9a8c\u3002\u8fd9\u79cd\u4e0d\u7a33\u5b9a\u884c\u4e3a\u65e2\u964d\u4f4e\u4e86\u4efb\u52a1\u7684\u6210\u529f\u7387\uff0c\u8fd8\u963b\u788d\u4e86\u667a\u80fd\u4f53\u5728\u771f\u5b9e\u73af\u5883\u4e2d\u7684\u90e8\u7f72\u4e0e\u5e94\u7528\u3002<\/p>\n\n\n\n<p>\u4e3a\u4e86\u5f25\u5408\u201c\u77e5\u8bc6-\u6267\u884c\u201d\u5dee\u8ddd\uff0c\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86 UI-Evol \u65b9\u6cd5\u3002\u4f5c\u4e3a\u4e00\u4e2a\u5373\u63d2\u5373\u7528\u7684\u6a21\u5757\uff0cUI-Evol \u53ef\u7528\u4e8e\u81ea\u4e3b\u7684\u56fe\u5f62\u7528\u6237\u754c\u9762\uff08GUI\uff09\u77e5\u8bc6\u8fdb\u5316\uff0c\u4ece\u800c\u63d0\u5347 Computer Use Agent \u5728\u771f\u5b9e\u4e16\u754c\u4efb\u52a1\u4e2d\u7684\u6267\u884c\u529b\u4e0e\u53ef\u9760\u6027\u3002\u5728 OSWorld \u6d4b\u8bd5\u57fa\u51c6\u4e0a\uff0cUI-Evol \u4ee5\u540c\u7b49\u7684\u57fa\u7840\u6a21\u578b\u5237\u65b0\u4e86\u6700\u9ad8\u6210\u529f\u7387\u3002\u76f8\u5173\u8bba\u6587\u5df2\u88ab ICML 2025 Computer Use Workshop \u63a5\u6536\u3002<\/p>\n\n\n\n<p><em><strong>UI-Evol: Automatic Knowledge Evolving for Computer Use Agents<\/strong><\/em><\/p>\n\n\n\n<p><em>\u8bba\u6587\u94fe\u63a5\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" rel=\"noopener noreferrer\" target=\"_blank\" href=\"https:\/\/arxiv.org\/abs\/2505.21964\">https:\/\/arxiv.org\/abs\/2505.21964<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/em><\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"452\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-1-1024x452.png\" alt=\"graphical user interface, text, application\" class=\"wp-image-1143912\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-1-1024x452.png 1024w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-1-300x133.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-1-768x339.png 768w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-1-1536x679.png 1536w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-1-240x106.png 240w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-1.png 1874w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\">\u56fe1\uff1a\u5c3d\u7ba1\u5916\u90e8\u77e5\u8bc6\u7406\u8bba\u4e0a\u662f\u6b63\u786e\u7684\uff0c\u4f46\u5e76\u4e0d\u80fd\u88ab\u6709\u6548\u5229\u7528\u5230\u5b9e\u9645\u73af\u5883\u4e2d\uff08\u5de6\u4e0a\u548c\u53f3\u4e0a\u90e8\u5206\uff09\u3002\u7ecf\u8fc7 UI-Evol \u540e\u7684\u77e5\u8bc6\u66f4\u52a0\u7b26\u5408\u5b9e\u9645\u73af\u5883\uff08\u5de6\u4e0b\u548c\u53f3\u4e0b\u90e8\u5206\uff09\u3002<\/figcaption><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"\u56de\u6eaf\u4e0e\u6279\u5224-\u77e5\u8bc6\u8fdb\u5316\u7684\u4e24\u4e2a\u5173\u952e\u9636\u6bb5\">\u56de\u6eaf\u4e0e\u6279\u5224\uff1a\u77e5\u8bc6\u8fdb\u5316\u7684\u4e24\u4e2a\u5173\u952e\u9636\u6bb5<\/h2>\n\n\n\n<p>UI-Evol \u6a21\u5757\u5305\u542b\u4e24\u4e2a\u6838\u5fc3\u9636\u6bb5\u3002\u7b2c\u4e00\u4e2a\u9636\u6bb5\u662f\u201c\u56de\u6eaf\u9636\u6bb5\u201d\uff08retrace stage\uff09\uff0c\u5176\u4e3b\u8981\u529f\u80fd\u662f\u4ece\u5b9e\u9645\u7684\u667a\u80fd\u4f53\u4e0e\u73af\u5883\u4ea4\u4e92\u4e2d\u63d0\u53d6\u201c\u5fe0\u5b9e\u5ba2\u89c2\u7684\u52a8\u4f5c\u5e8f\u5217\u201d\u3002\u8fd9\u610f\u5473\u7740 UI-Evol \u80fd\u591f\u8bb0\u5f55\u5e76\u7406\u89e3\u667a\u80fd\u4f53\u5728\u6267\u884c\u4efb\u52a1\u65f6\u6240\u91c7\u53d6\u7684\u5177\u4f53\u3001\u6709\u6548\u7684\u6b65\u9aa4\u3002<\/p>\n\n\n\n<p>\u7b2c\u4e8c\u4e2a\u9636\u6bb5\u662f\u201c\u6279\u5224\u9636\u6bb5\u201d\uff08critique stage\uff09\u3002\u5728\u6b64\u9636\u6bb5\uff0c\u7cfb\u7edf\u4f1a\u5c06\u56de\u6eaf\u9636\u6bb5\u63d0\u53d6\u51fa\u7684\u52a8\u4f5c\u5e8f\u5217\u4e0e\u5916\u90e8\u53c2\u8003\u8fdb\u884c\u6bd4\u5bf9\u5206\u6790\uff0c\u4ece\u800c\u4fee\u6b63\u3001\u5b8c\u5584\u73b0\u6709\u77e5\u8bc6\u3002\u8fd9\u79cd\u201c\u56de\u6eaf+\u6279\u5224\u201d\u7684\u4e24\u9636\u6bb5\u65b9\u6cd5\u53ef\u4ee5\u786e\u4fdd\u667a\u80fd\u4f53\u6240\u83b7\u5f97\u7684\u77e5\u8bc6\u4e0d\u4ec5\u662f\u7406\u8bba\u4e0a\u6b63\u786e\u7684\uff0c\u800c\u4e14\u80fd\u591f\u5728\u5b9e\u9645\u4ea4\u4e92\u4e2d\u6709\u6548\u6267\u884c\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"544\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-2-1024x544.png\" alt=\"graphical user interface, diagram\" class=\"wp-image-1143913\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-2-1024x544.png 1024w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-2-300x159.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-2-768x408.png 768w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-2-1536x817.png 1536w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-2-240x128.png 240w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-2.png 1710w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\">\u56fe2\uff1aUI-Evol \u5206\u4e3a\u4e24\u4e2a\u9636\u6bb5\uff0c\u901a\u8fc7\u5b9e\u9645\u7684\u667a\u80fd\u4f53\u4e0e\u73af\u5883\u4ea4\u4e92\u7684\u8f68\u8ff9\u6765\u66f4\u65b0\u5916\u90e8\u77e5\u8bc6\uff0c\u4f7f\u5176\u7b26\u5408\u5b9e\u9645\u4ea4\u4e92\u73af\u5883\u3002<\/figcaption><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"\u5b9e\u9a8c\u9a8c\u8bc1-\u8ba9\u667a\u80fd\u4f53\u66f4\u7a33\u5b9a-\u66f4\u53ef\u9884\u6d4b\">\u5b9e\u9a8c\u9a8c\u8bc1\uff1a\u8ba9\u667a\u80fd\u4f53\u66f4\u7a33\u5b9a\u3001\u66f4\u53ef\u9884\u6d4b<\/h2>\n\n\n\n<p>\u7814\u7a76\u5458\u4eec\u5728 OSWorld \u57fa\u51c6\u6d4b\u8bd5\u4e0a\u5bf9 UI-Evol \u8fdb\u884c\u4e86\u5168\u9762\u7684\u5b9e\u9a8c\uff0c\u5e76\u4f7f\u7528\u4e86\u6700\u5148\u8fdb\u7684\u667a\u80fd\u4f53 Agent S2\u3002\u5b9e\u9a8c\u7ed3\u679c\u663e\u793a\uff0cUI-Evol \u65e2\u663e\u8457\u63d0\u5347\u4e86\u4efb\u52a1\u6027\u80fd\uff0c\u66f4\u91cd\u8981\u7684\u662f\uff0c\u5b83\u8fd8\u89e3\u51b3\u4e86\u4e00\u4e2a\u4e4b\u524d\u88ab\u5ffd\u89c6\u7684\u95ee\u9898\u2014\u2014Computer Use Agent \u5b58\u5728\u201c\u9ad8\u884c\u4e3a\u6807\u51c6\u5dee\u201d\u3002\u8fd9\u610f\u5473\u7740\u5728 UI-Evol \u7684\u5e2e\u52a9\u4e0b\uff0c\u667a\u80fd\u4f53\u5728\u6267\u884c\u4efb\u52a1\u65f6\u7684\u884c\u4e3a\u6a21\u5f0f\u53d8\u5f97\u66f4\u52a0\u7a33\u5b9a\u4e14\u53ef\u9884\u6d4b\u3002<\/p>\n\n\n\n<p>\u7531\u6b64\u53ef\u89c1\uff0cUI-Evol \u5927\u5e45\u63d0\u9ad8\u4e86\u667a\u80fd\u4f53\u7684\u53ef\u9760\u6027\uff0c\u5e26\u6765\u4e86\u8ba1\u7b97\u673a\u4f7f\u7528\u4efb\u52a1\u4e0a\u5353\u8d8a\u7684\u6027\u80fd\u8868\u73b0\u3002\u8fd9\u4e9b\u6210\u679c\u9a8c\u8bc1\u4e86 UI-Evol \u5728\u63d0\u5347\u667a\u80fd\u4f53\u5b9e\u9645\u5e94\u7528\u80fd\u529b\u548c\u7a33\u5b9a\u6027\u65b9\u9762\u7684\u6709\u6548\u6027\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><img loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"349\" src=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-3-1024x349.png\" alt=\"table\" class=\"wp-image-1143914\" srcset=\"https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-3-1024x349.png 1024w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-3-300x102.png 300w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-3-768x262.png 768w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-3-1536x523.png 1536w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-3-240x82.png 240w, https:\/\/www.microsoft.com\/en-us\/research\/wp-content\/uploads\/2025\/07\/ui-evol-3.png 1629w\" sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><figcaption class=\"wp-element-caption\">\u88681\uff1a\u5728\u590d\u73b0\u5b9e\u9a8c\u4e2d\uff0c\u57fa\u4e8e GPT-4o \u548c OpenAI-o3 \u7684 Computer Use Agent \u51fa\u73b0\u4e86\u8f83\u5927\u7684\u968f\u673a\u6027\u3002\u52a0\u5165\u7ecf\u8fc7 UI-Evol \u540e\u7684\u77e5\u8bc6\u4e0d\u4ec5\u63d0\u9ad8\u4e86\u6027\u80fd\uff0c\u800c\u4e14\u63d0\u9ad8\u4e86\u7a33\u5b9a\u6027\u548c\u53ef\u590d\u73b0\u6027\u3002<\/figcaption><\/figure>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"\u4e3a\u667a\u80fd\u4f53\u6253\u5f00-\u7528\u5f97\u4e0a-\u7684\u65b0\u5c40\u9762\">\u4e3a\u667a\u80fd\u4f53\u6253\u5f00\u201c\u7528\u5f97\u4e0a\u201d\u7684\u65b0\u5c40\u9762<\/h2>\n\n\n\n<p>UI-Evol \u7684\u63d0\u51fa\uff0c\u89e3\u51b3\u4e86\u957f\u671f\u56f0\u6270\u667a\u80fd\u4f53\u4efb\u52a1\u6267\u884c\u4e2d\u201c\u77e5\u8bc6-\u884c\u52a8\u201d\u8131\u8282\u7684\u95ee\u9898\uff0c\u4e3a\u63d0\u5347 AI \u5728\u771f\u5b9e\u4e16\u754c\u4e2d\u7684\u5b9e\u7528\u6027\u63d0\u4f9b\u4e86\u6709\u6548\u8def\u5f84\u3002\u672a\u6765\uff0c\u968f\u7740\u667a\u80fd\u4f53\u5728\u529e\u516c\u81ea\u52a8\u5316\u3001\u865a\u62df\u52a9\u7406\u3001\u8f6f\u4ef6\u64cd\u63a7\u7b49\u573a\u666f\u4e2d\u7684\u6df1\u5165\u5e94\u7528\uff0cUI-Evol \u6709\u671b\u6210\u4e3a\u5176\u7a33\u5b9a\u6027\u4e0e\u53ef\u9760\u6027\u7684\u91cd\u8981\u4fdd\u969c\u6a21\u5757\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u7f16\u8005\u6309\uff1a\u5f53 AI \u667a\u80fd\u4f53\u8d70\u51fa\u5b9e\u9a8c\u5ba4\uff0c\u5728\u771f\u5b9e\u4e16\u754c\u4e2d\u6267\u884c\u4efb\u52a1\u65f6\uff0c\u53ea\u6709\u77e5\u8bc6\u8fd8\u8fdc\u8fdc\u4e0d\u591f\u3002\u4f46\u5982\u4eca\u7684\u8bb8\u591a\u667a\u80fd\u4f53\u5373\u4f7f\u62e5\u6709\u6b63\u786e\u7684\u4fe1\u606f\uff0c\u53ef\u5728\u5b9e\u9645\u64cd\u4f5c\u4e2d\u4ecd\u9891\u9891\u5931\u8bef\uff0c\u4efb\u52a1\u5b8c\u6210\u7387\u4ee4\u4eba\u582a\u5fe7\u3002\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u5bf9\u6b64\u63d0\u51fa\u4e86\u4e00\u4e2a\u65b0\u65b9\u6cd5 UI-Evol\uff0c\u65e8\u5728\u5e2e\u52a9\u8ba1\u7b97\u673a\u4f7f\u7528\u667a\u80fd\u4f53\uff08Computer Use Agent\uff09\u8fdb\u5316\u77e5\u8bc6\uff0c\u4f7f\u667a\u80fd\u4f53\u4e0d\u4ec5\u201c\u77e5\u9053\u8be5\u600e\u4e48\u505a\u201d\uff0c\u8fd8\u8981\u771f\u6b63\u201c\u505a\u5f97\u5230\u201d\u3002\u5728\u8be5\u65b9\u6cd5\u7684\u52a0\u6301\u4e0b\uff0c\u667a\u80fd\u4f53\u5728\u5b9e\u9645\u6267\u884c\u80fd\u529b\u3001\u7a33\u5b9a\u6027\u548c\u53ef\u9760\u6027\u65b9\u9762\u5747\u5f97\u5230\u663e\u8457\u63d0\u5347\u3002\u76f8\u5173\u8bba\u6587\u88ab ICML 2025 Computer Use Workshop \u63a5\u6536\u3002 \u5728\u5f53\u524d\u8ba1\u7b97\u673a\u4f7f\u7528\u667a\u80fd\u4f53\uff08Computer Use Agent\uff09\u7684\u5f00\u53d1\u4e2d\uff0c\u5916\u90e8\u77e5\u8bc6\u626e\u6f14\u7740\u81f3\u5173\u91cd\u8981\u7684\u89d2\u8272\u3002\u7136\u800c\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u53d1\u73b0\u4e86\u4e00\u4e2a\u5173\u952e\u7684\u201c\u77e5\u8bc6-\u6267\u884c\u201d\u9e3f\u6c9f\uff0c\u5373\u6240\u68c0\u7d22\u5230\u7684\u77e5\u8bc6\u5f80\u5f80\u672a\u80fd\u6709\u6548\u5730\u8f6c\u5316\u4e3a\u5b9e\u9645\u7684\u4efb\u52a1\u6267\u884c\u3002\u8c03\u67e5\u5206\u6790\u663e\u793a\uff0c\u5373\u4f7f\u63d0\u4f9b\u7ed9\u667a\u80fd\u4f53\uff08agent\uff0990%\u7684\u6b63\u786e\u77e5\u8bc6\uff0c\u5176\u6267\u884c\u4efb\u52a1\u7684\u6210\u529f\u7387\u4e5f\u53ea\u670941%\u3002\u8fd9\u4e00\u5dee\u8ddd\u66b4\u9732\u51fa\u4e00\u4e2a\u6df1\u5c42\u95ee\u9898\uff1a\u5c3d\u7ba1\u667a\u80fd\u4f53\u62e5\u6709\u76f8\u5173\u77e5\u8bc6\uff0c\u4f46\u5728\u771f\u5b9e\u7684\u8ba1\u7b97\u673a\u4f7f\u7528\u73af\u5883\u4e2d\uff0c\u8fd9\u4e9b\u77e5\u8bc6\u5374\u96be\u4ee5\u88ab\u6709\u6548\u5e94\u7528\uff0c\u5bfc\u81f4\u4efb\u52a1\u5b8c\u6210\u7387\u4f4e\u4e0b\uff0c\u77e5\u8bc6\u4e0e\u884c\u52a8\u4e4b\u95f4\u5b58\u5728\u660e\u663e\u7684\u8131\u8282\u3002\u4e5f\u5c31\u662f\u8bf4\uff0c\u667a\u80fd\u4f53\u867d\u201c\u77e5\u5176\u7136\u201d\uff0c\u5374\u672a\u5fc5\u80fd\u201c\u884c\u5176\u662f\u201d\u3002 \u901a\u8fc7\u5b9e\u9a8c\uff0c\u7814\u7a76\u5458\u4eec\u8fdb\u4e00\u6b65\u53d1\u73b0\uff0c\u73b0\u9636\u6bb5 Computer Use Agent \u6267\u884c\u4efb\u52a1\u7684\u6210\u529f\u7387\u5b58\u5728\u8f83\u5f3a\u7684\u968f\u673a\u6027\uff0c\u4e25\u91cd\u5f71\u54cd\u7528\u6237\u7684\u5b9e\u9645\u4f53\u9a8c\u3002\u8fd9\u79cd\u4e0d\u7a33\u5b9a\u884c\u4e3a\u65e2\u964d\u4f4e\u4e86\u4efb\u52a1\u7684\u6210\u529f\u7387\uff0c\u8fd8\u963b\u788d\u4e86\u667a\u80fd\u4f53\u5728\u771f\u5b9e\u73af\u5883\u4e2d\u7684\u90e8\u7f72\u4e0e\u5e94\u7528\u3002 \u4e3a\u4e86\u5f25\u5408\u201c\u77e5\u8bc6-\u6267\u884c\u201d\u5dee\u8ddd\uff0c\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86 UI-Evol \u65b9\u6cd5\u3002\u4f5c\u4e3a\u4e00\u4e2a\u5373\u63d2\u5373\u7528\u7684\u6a21\u5757\uff0cUI-Evol \u53ef\u7528\u4e8e\u81ea\u4e3b\u7684\u56fe\u5f62\u7528\u6237\u754c\u9762\uff08GUI\uff09\u77e5\u8bc6\u8fdb\u5316\uff0c\u4ece\u800c\u63d0\u5347 Computer Use Agent \u5728\u771f\u5b9e\u4e16\u754c\u4efb\u52a1\u4e2d\u7684\u6267\u884c\u529b\u4e0e\u53ef\u9760\u6027\u3002\u5728 OSWorld \u6d4b\u8bd5\u57fa\u51c6\u4e0a\uff0cUI-Evol \u4ee5\u540c\u7b49\u7684\u57fa\u7840\u6a21\u578b\u5237\u65b0\u4e86\u6700\u9ad8\u6210\u529f\u7387\u3002\u76f8\u5173\u8bba\u6587\u5df2\u88ab ICML 2025 Computer Use Workshop \u63a5\u6536\u3002 UI-Evol: Automatic Knowledge Evolving for Computer Use Agents \u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/abs\/2505.21964 (opens in new tab) UI-Evol \u6a21\u5757\u5305\u542b\u4e24\u4e2a\u6838\u5fc3\u9636\u6bb5\u3002\u7b2c\u4e00\u4e2a\u9636\u6bb5\u662f\u201c\u56de\u6eaf\u9636\u6bb5\u201d\uff08retrace stage\uff09\uff0c\u5176\u4e3b\u8981\u529f\u80fd\u662f\u4ece\u5b9e\u9645\u7684\u667a\u80fd\u4f53\u4e0e\u73af\u5883\u4ea4\u4e92\u4e2d\u63d0\u53d6\u201c\u5fe0\u5b9e\u5ba2\u89c2\u7684\u52a8\u4f5c\u5e8f\u5217\u201d\u3002\u8fd9\u610f\u5473\u7740 UI-Evol \u80fd\u591f\u8bb0\u5f55\u5e76\u7406\u89e3\u667a\u80fd\u4f53\u5728\u6267\u884c\u4efb\u52a1\u65f6\u6240\u91c7\u53d6\u7684\u5177\u4f53\u3001\u6709\u6548\u7684\u6b65\u9aa4\u3002 [&hellip;]<\/p>\n","protected":false},"author":34512,"featured_media":1143915,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr-content-parent":1012650,"msr_hide_image_in_river":null,"footnotes":""},"research-area":[13556],"msr-locale":[268881],"msr-post-option":[],"class_list":["post-1143911","msr-blog-post","type-msr-blog-post","status-publish","has-post-thumbnail","hentry","msr-research-area-artificial-intelligence","msr-locale-zh_cn"],"msr_assoc_parent":{"id":1012650,"type":"lab"},"_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1143911","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-blog-post"}],"author":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/users\/34512"}],"version-history":[{"count":2,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1143911\/revisions"}],"predecessor-version":[{"id":1143917,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1143911\/revisions\/1143917"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media\/1143915"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1143911"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1143911"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1143911"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1143911"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}