{"id":1075818,"date":"2024-05-08T09:58:02","date_gmt":"2024-05-08T16:58:02","guid":{"rendered":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/?post_type=msr-blog-post&#038;p=1075818"},"modified":"2024-09-25T04:17:59","modified_gmt":"2024-09-25T11:17:59","slug":"new-arrival-in-research-11","status":"publish","type":"msr-blog-post","link":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/articles\/new-arrival-in-research-11\/","title":{"rendered":"ICLR\u4e0a\u65b0 | \u5f3a\u5316\u5b66\u4e60\u3001\u6269\u6563\u6a21\u578b\u3001\u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b\uff0c\u4f60\u60f3\u4e86\u89e3\u7684\u524d\u6cbf\u65b9\u5411\u8fdb\u5c55\u5168\u90fd\u6709"},"content":{"rendered":"\n<p>\u7f16\u8005\u6309\uff1a\u6b22\u8fce\u9605\u8bfb\u201c\u79d1\u7814\u4e0a\u65b0\u201d\u680f\u76ee\uff01\u201c\u79d1\u7814\u4e0a\u65b0\u201d\u6c47\u805a\u4e86\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u6700\u65b0\u7684\u521b\u65b0\u6210\u679c\u4e0e\u79d1\u7814\u52a8\u6001\u3002\u5728\u8fd9\u91cc\uff0c\u4f60\u53ef\u4ee5\u5feb\u901f\u6d4f\u89c8\u7814\u7a76\u9662\u7684\u4eae\u70b9\u8d44\u8baf\uff0c\u4fdd\u6301\u5bf9\u524d\u6cbf\u9886\u57df\u7684\u654f\u9510\u55c5\u89c9\uff0c\u540c\u65f6\u4e5f\u80fd\u627e\u5230\u5148\u8fdb\u5b9e\u7528\u7684\u5f00\u6e90\u5de5\u5177\u3002<\/p>\n\n\n\n<p>\u672c\u5468\uff0c\u5168\u7403\u6700\u8d1f\u76db\u540d\u7684\u4eba\u5de5\u667a\u80fd\u76db\u4f1a\u4e4b\u4e00 ICLR \u5927\u4f1a\u5c06\u5728\u5965\u5730\u5229\u7ef4\u4e5f\u7eb3\u4e3e\u529e\u3002\u6240\u4ee5\uff0c\u4eca\u5929\u7684\u201c\u79d1\u7814\u4e0a\u65b0\u201d\u5c06\u4e3a\u5927\u5bb6\u5e26\u6765\u591a\u7bc7\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u5728 ICLR 2024 \u4e0a\u7684\u7cbe\u9009\u8bba\u6587\u89e3\u8bfb\uff0c\u6d89\u53ca\u9886\u57df\u6db5\u76d6\u6df1\u5ea6\u5f3a\u5316\u5b66\u4e60\u3001\u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b\u3001\u65f6\u95f4\u5e8f\u5217\u6269\u6563\u6a21\u578b\u3001\u65e0\u76d1\u7763\u5b66\u4e60\u7b49\u591a\u4e2a\u524d\u6cbf\u4e3b\u9898\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"\u5e94\u5bf9\u6df1\u5ea6\u5f3a\u5316\u5b66\u4e60\u4e2d\u7684\u4fe1\u53f7\u5ef6\u8fdf\u95ee\u9898\">\u5e94\u5bf9\u6df1\u5ea6\u5f3a\u5316\u5b66\u4e60\u4e2d\u7684\u4fe1\u53f7\u5ef6\u8fdf\u95ee\u9898<\/h3>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"1416\" height=\"350\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-1.png\" alt=\"Addressing Signal Delay in Deep Reinforcement Learning\n\" class=\"wp-image-1076742\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-1.png 1416w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-1-300x74.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-1-1024x253.png 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-1-768x190.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-1-240x59.png 240w\" sizes=\"auto, (max-width: 1416px) 100vw, 1416px\" \/><\/figure>\n\n\n\n<p>\u8bba\u6587\u94fe\u63a5\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/openreview.net\/forum?id=Z8UfDs4J46\" target=\"_blank\" rel=\"noopener noreferrer\">https:\/\/openreview.net\/forum?id=Z8UfDs4J46<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>\u8fd1\u5e74\u6765\uff0c\u6df1\u5ea6\u5f3a\u5316\u5b66\u4e60\uff08DRL\uff09\u53ca\u5176\u5e94\u7528\u8fc5\u901f\u53d1\u5c55\uff0c\u5b83\u4e0d\u4ec5\u5728\u865a\u62df\u4efb\u52a1\uff08\u5982\u89c6\u9891\u6e38\u620f\u548c\u6a21\u62df\u673a\u5668\u4eba\u73af\u5883\uff09\u4e0a\u53d6\u5f97\u4e86\u6210\u529f\uff0c\u4e5f\u5728\u8bb8\u591a\u5177\u6709\u6311\u6218\u6027\u7684\u73b0\u5b9e\u4e16\u754c\u4efb\u52a1\u4e2d\u5f97\u5230\u4e86\u8bc1\u660e\uff0c\u4f8b\u5982\u63a7\u5236\u6258\u5361\u9a6c\u514b\u548c\u901a\u8fc7\u4eba\u7c7b\u53cd\u9988\u8c03\u6574\u5927\u8bed\u8a00\u6a21\u578b\u3002\u7136\u800c\uff0c\u5bfc\u81f4\u667a\u80fd\u4f53\u53ef\u80fd\u65e0\u6cd5\u7acb\u5373\u89c2\u5bdf\u5230\u5f53\u524d\u73af\u5883\u72b6\u6001\u6216\u5176\u884c\u52a8\u65e0\u6cd5\u7acb\u5373\u5f71\u54cd\u73af\u5883\u7684\u4fe1\u53f7\u5ef6\u8fdf\uff0c\u5728\u6df1\u5ea6\u5f3a\u5316\u5b66\u4e60\u7814\u7a76\u4e2d\u957f\u671f\u5b58\u5728\u4e14\u7ecf\u5e38\u88ab\u5ffd\u89c6\u3002\u8be5\u95ee\u9898\u5e7f\u6cdb\u5b58\u5728\u4e8e\u5404\u79cd\u5b9e\u9645\u5e94\u7528\u4e2d\uff0c\u5bf9\u57fa\u4e8e\u6df1\u5ea6\u5f3a\u5316\u5b66\u4e60\u89e3\u51b3\u65b9\u6848\u7684\u6709\u6548\u6027\u4ea7\u751f\u4e86\u91cd\u5927\u5f71\u54cd\uff0c\u56e0\u6b64\u8be5\u6311\u6218\u8feb\u5207\u9700\u8981\u7814\u7a76\u8fdb\u884c\u5e94\u5bf9\u3002<\/p>\n\n\n\n<p>\u4e3a\u4e86\u89e3\u51b3 DRL \u4e2d\u7684\u4fe1\u53f7\u5ef6\u8fdf\u95ee\u9898\uff0c\u7814\u7a76\u5458\u4eec\u9996\u5148\u901a\u8fc7\u6269\u5c55\u9a6c\u5c14\u53ef\u592b\u51b3\u7b56\u8fc7\u7a0b\u6846\u67b6\u6765\u5b9a\u4e49\u5ef6\u8fdf\u89c2\u6d4b\u9a6c\u5c14\u53ef\u592b\u51b3\u7b56\u8fc7\u7a0b\uff08DOMDP\uff09\uff0c\u4ece\u800c\u5c06\u4fe1\u53f7\u5ef6\u8fdf\u7684\u60c5\u51b5\u7eb3\u5165\u8003\u8651\u4e4b\u4e2d\u3002\u7136\u540e\uff0c\u7814\u7a76\u5458\u4eec\u5728\u8bba\u6587\u4e2d\u9610\u660e\u4e86 DRL \u91cc\u4fe1\u53f7\u5ef6\u8fdf\u5b58\u5728\u7684\u6311\u6218\uff0c\u5e76\u5c55\u793a\u4e86\u5e38\u89c4 DRL \u7b97\u6cd5\u548c\u90e8\u5206\u53ef\u89c2\u6d4b\u9a6c\u5c14\u53ef\u592b\u51b3\u7b56\u8fc7\u7a0b\uff08POMDP\uff09\u7684\u901a\u7528\u65b9\u6cd5\u53d7\u5230\u5ef6\u8fdf\u7684\u4e25\u91cd\u5f71\u54cd\u3002<\/p>\n\n\n\n<p>\u9488\u5bf9\u8fd9\u4e9b\u6311\u6218\uff0c\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u4e00\u7cfb\u5217\u65b0\u65b9\u6cd5\uff0c\u65e8\u5728\u63d0\u9ad8\u5b58\u5728\u5ef6\u8fdf\u65f6 DRL \u7b97\u6cd5\u7684\u6027\u80fd\u3002\u7ed3\u5408\u7406\u8bba\u89c1\u89e3\u548c\u5b9e\u9645\u7b97\u6cd5\u8c03\u6574\uff0c\u7814\u7a76\u5458\u4eec\u6269\u5c55\u4e86\u4f20\u7edf\u7684 actor-critic \u6846\u67b6\uff0c\u5e76\u63d0\u51fa\u4e86\u6709\u6548\u7684\u7b56\u7565\u6765\u514b\u670d\u8fd9\u4e9b\u6311\u6218\u3002\u5145\u5206\u7684\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u5728\u5177\u6709\u8f83\u5927\u5ef6\u8fdf\u7684\u8fde\u7eed\u673a\u5668\u4eba\u63a7\u5236\u4efb\u52a1\u4e2d\uff0c\u91c7\u7528\u8be5\u8bba\u6587\u63d0\u51fa\u7684\u65b9\u6cd5\u540e\uff0cDRL \u7b97\u6cd5\u53d6\u5f97\u4e86\u5353\u8d8a\u7684\u6027\u80fd\uff0c\u5176\u7ed3\u679c\u4e0e\u65e0\u5ef6\u8fdf\u60c5\u51b5\u76f8\u6bd4\uff0c\u6027\u80fd\u635f\u5931\u8f83\u5c0f\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"624\" height=\"177\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-2.png\" alt=\"Addressing Signal Delay in Deep Reinforcement Learning | four charts comparing delay time\" class=\"wp-image-1075827\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-2.png 624w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-2-300x85.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-2-240x68.png 240w\" sizes=\"auto, (max-width: 624px) 100vw, 624px\" \/><figcaption class=\"wp-element-caption\">\u56fe1\uff1a\u8be5\u8bba\u6587\u4e2d\u7684\u65b9\u6cd5\uff08\u7ea2\u8272\u865a\u7ebf\uff09\u53ef\u4ee5\u5728\u6709\u4fe1\u53f7\u5ef6\u8fdf\u7684\u60c5\u51b5\u4e0b\u4fdd\u6301\u8f83\u597d\u7684\u6548\u679c\uff0c\u800c\u5176\u4ed6\u5e38\u7528\u7684\u65b9\u6cd5\u5728\u6709\u5ef6\u8fdf\u7684\u60c5\u51b5\u4e0b\u8868\u73b0\u663e\u8457\u4e0b\u964d\uff08\u4f5c\u4e3a\u5bf9\u6bd4\uff0c\u9ed1\u8272\u865a\u7ebf\u662f\u6ca1\u6709\u4fe1\u53f7\u5ef6\u8fdf\u60c5\u51b5\u4e0b\u7684\u8868\u73b0\uff09\u3002<\/figcaption><\/figure>\n\n\n\n<p>\u8fd9\u9879\u7814\u7a76\u5728\u89e3\u51b3 DRL \u4e2d\u4e00\u4e2a\u57fa\u672c\u6311\u6218\u65b9\u9762\u8fc8\u51fa\u4e86\u91cd\u8981\u7684\u4e00\u6b65\uff0c\u4e0d\u4ec5\u62d3\u5bbd\u4e86\u5176\u5728\u73b0\u5b9e\u73af\u5883\u4e2d\u7684\u5e94\u7528\u8303\u56f4\uff0c\u4e5f\u4e3a\u81ea\u4e3b\u7cfb\u7edf\u7684\u6301\u7eed\u53d1\u5c55\u505a\u51fa\u4e86\u8d21\u732e\u3002\u901a\u8fc7\u5f00\u53d1\u6709\u6548\u5e94\u5bf9\u4fe1\u53f7\u5ef6\u8fdf\u7684\u65b9\u6cd5\uff0c\u7814\u7a76\u5458\u4eec\u589e\u5f3a\u4e86 DRL \u7684\u5b9e\u7528\u6027\u548c\u53ef\u9760\u6027\uff0c\u4e3a\u5176\u5728\u975e\u7406\u60f3\u6761\u4ef6\u4e0b\u7684\u5e94\u7528\u5960\u5b9a\u4e86\u57fa\u7840\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"research2\">\u7ea7\u8054\u5f3a\u5316\u5b66\u4e60<\/h3>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"537\" height=\"219\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-3.png\" alt=\"Cascade Reinforcement Learning\" class=\"wp-image-1075830\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-3.png 537w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-3-300x122.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-3-240x98.png 240w\" sizes=\"auto, (max-width: 537px) 100vw, 537px\" \/><\/figure>\n\n\n\n<p>\u8bba\u6587\u94fe\u63a5\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/arxiv.org\/abs\/2401.08961\" target=\"_blank\" rel=\"noopener noreferrer\">https:\/\/arxiv.org\/abs\/2401.08961<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>\u8fd1\u5e74\u6765\uff0c\u4e00\u79cd\u540d\u4e3a\u7ea7\u8054\u591a\u81c2\u8001\u864e\u673a\uff08cascading bandits\uff09[Kveton et al., 2015]\u7684\u6a21\u578b\u53d7\u5230\u4e86\u5e7f\u6cdb\u5173\u6ce8\uff0c\u5728\u63a8\u8350\u7cfb\u7edf\u3001\u5728\u7ebf\u5e7f\u544a\u4e2d\u5e94\u7528\u5e7f\u6cdb\u3002\u5728\u7ea7\u8054\u591a\u81c2\u8001\u864e\u673a\u4e2d\uff0c\u667a\u80fd\u4f53\u9700\u8981\u5728\u4f17\u591a\u9009\u9879\u4e2d\u6311\u9009\u4e00\u4e2a\u9009\u9879\u5217\u8868\u63a8\u8350\u7ed9\u7528\u6237\uff0c\u6bcf\u4e2a\u9009\u9879\u90fd\u6709\u4e00\u4e2a\u672a\u77e5\u7684\u5438\u5f15\u6982\u7387\u3002\u667a\u80fd\u4f53\u7684\u76ee\u6807\u5219\u662f\u4e0d\u65ad\u4f18\u5316\u63a8\u8350\u7684\u9009\u9879\u5217\u8868\uff0c\u4ee5\u6700\u5927\u5316\u671f\u671b\u7d2f\u79ef\u5956\u52b1\uff08\u70b9\u51fb\u7387\uff09\u3002<\/p>\n\n\n\n<p>\u7136\u800c\uff0c\u73b0\u6709\u7684\u7ea7\u8054\u591a\u81c2\u8001\u864e\u673a\u6a21\u578b\u5ffd\u7565\u4e86\u7528\u6237\u72b6\u6001\uff08\u5982\u7528\u6237\u5386\u53f2\u884c\u4e3a\uff09\u5bf9\u63a8\u8350\u7684\u5f71\u54cd\uff0c\u4ee5\u53ca\u7528\u6237\u72b6\u6001\u53ef\u80fd\u7684\u6539\u53d8\u3002\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e00\u95ee\u9898\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u540d\u4e3a\u7ea7\u8054\u5f3a\u5316\u5b66\u4e60\uff08cascading reinforcement learning\uff09\u7684\u6a21\u578b\u3002\u5728\u8be5\u6a21\u578b\u4e2d\uff0c\u6bcf\u4e2a\u7528\u6237\u72b6\u6001-\u9009\u9879\u7684\u5339\u914d\u5bf9\u6709\u4e00\u4e2a\u672a\u77e5\u7684\u5438\u5f15\u6982\u7387\u3001\u4e00\u4e2a\u672a\u77e5\u7684\u72b6\u6001\u8f6c\u79fb\u5206\u5e03\u548c\u4e00\u4e2a\u5956\u52b1\u3002\u5982\u56fe2\u6240\u793a\uff0c\u5728\u6bcf\u4e2a\u65f6\u523b\uff0c\u667a\u80fd\u4f53\u4f1a\u5148\u89c2\u5bdf\u5230\u5f53\u524d\u7684\u7528\u6237\u72b6\u6001\uff0c\u7136\u540e\u63a8\u8350\u4e00\u4e2a\u957f\u5ea6\u4e3a m \u7684\u9009\u9879\u5217\u8868\u3002\u5982\u679c\u7528\u6237\u5bf9\u67d0\u4e00\u9009\u9879\u611f\u5174\u8da3\u5e76\u70b9\u51fb\uff0c\u90a3\u4e48\u7528\u6237\u5c06\u8f6c\u79fb\u5230\u4e0b\u4e00\u72b6\u6001\uff0c\u667a\u80fd\u4f53\u5219\u4f1a\u83b7\u5f97\u4e00\u4e2a\u5956\u52b1\u3002\u667a\u80fd\u4f53\u7684\u76ee\u6807\u662f\u6700\u5927\u5316\u671f\u671b\u7d2f\u79ef\u5956\u52b1\uff0c\u56e0\u6b64\u8be5\u6a21\u578b\u80fd\u6709\u6548\u5730\u5c06\u7528\u6237\u72b6\u6001\u53ca\u5176\u53d8\u5316\u7eb3\u5165\u63a8\u8350\u8fc7\u7a0b\u4e2d\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-medium\"><img loading=\"lazy\" decoding=\"async\" width=\"187\" height=\"300\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-4-187x300.png\" alt=\"Cascade Reinforcement Learning | diagram\" class=\"wp-image-1075833\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-4-187x300.png 187w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-4-112x180.png 112w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-4.png 283w\" sizes=\"auto, (max-width: 187px) 100vw, 187px\" \/><figcaption class=\"wp-element-caption\">\u56fe2\uff1a\u7ea7\u8054\u5f3a\u5316\u5b66\u4e60\u6a21\u578b<\/figcaption><\/figure>\n\n\n\n<p>\u9488\u5bf9\u8be5\u6a21\u578b\uff0c\u7814\u7a76\u5458\u4eec\u9996\u5148\u57fa\u4e8e\u52a8\u6001\u89c4\u5212\u8bbe\u8ba1\u4e86\u4e00\u4e2a\u5feb\u901f\u79bb\u7ebf\u6c42\u89e3\u5668 BestPerm\uff0c\u5b83\u80fd\u591f\u5728\u591a\u9879\u5f0f\u65f6\u95f4\u5185\u8ba1\u7b97\u51fa\u6700\u4f18\u7684\u9009\u9879\u5217\u8868\u3002\u7136\u540e\uff0c\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5 CascadingVI\uff0c\u8be5\u7b97\u6cd5\u80fd\u591f\u8fbe\u5230 O \u0303(H\u221aHSNK) \u7684\u540e\u6094\u5ea6\uff08regret\uff09\u4e0a\u754c\uff0c\u8fd9\u4e2a\u7ed3\u679c\u53ea\u4f9d\u8d56\u4e8e\u9009\u9879\u7684\u4e2a\u6570 \ud835\udc41, \u800c\u4e0e\u9009\u9879\u5217\u8868\u7684\u4e2a\u6570\uff08\u7ea6 N^m\uff09\u65e0\u5173\u3002\u56e0\u6b64\uff0c\u8be5\u7b97\u6cd5\u80fd\u540c\u65f6\u4fdd\u8bc1\u91c7\u6837\u548c\u8ba1\u7b97\u7684\u9ad8\u6548\u6027\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"research3\">DyVal\uff1a\u9996\u4e2a\u5927\u8bed\u8a00\u6a21\u578b\u7684\u52a8\u6001\u8bc4\u6d4b\u534f\u8bae<\/h3>\n\n\n\n<figure class=\"wp-block-image size-full is-resized\"><img loading=\"lazy\" decoding=\"async\" width=\"1431\" height=\"332\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-5.png\" alt=\"DyVal: Dynamic Evaluation of Large Language Models for Reasoning Tasks\n\" class=\"wp-image-1076748\" style=\"width:711px;height:auto\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-5.png 1431w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-5-300x70.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-5-1024x238.png 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-5-768x178.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-5-240x56.png 240w\" sizes=\"auto, (max-width: 1431px) 100vw, 1431px\" \/><\/figure>\n\n\n\n<p>\u8bba\u6587\u94fe\u63a5\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/arxiv.org\/abs\/2309.17167\" target=\"_blank\" rel=\"noopener noreferrer\">https:\/\/arxiv.org\/abs\/2309.17167<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>\u9879\u76ee\u94fe\u63a5\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/github.com\/microsoft\/promptbench\" target=\"_blank\" rel=\"noopener noreferrer\">https:\/\/github.com\/microsoft\/promptbench<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>DyVal 2 \u8bba\u6587\u94fe\u63a5\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/arxiv.org\/abs\/2402.14865\" target=\"_blank\" rel=\"noopener noreferrer\">https:\/\/arxiv.org\/abs\/2402.14865<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>\u5927\u8bed\u8a00\u6a21\u578b\uff08LLMs\uff09\u901a\u5e38\u90fd\u662f\u5728\u6d77\u91cf\u7684\u6570\u636e\u4e0a\u8fdb\u884c\u8bad\u7ec3\uff0c\u800c\u8fd9\u5c31\u5bfc\u81f4\u4e86\u6f5c\u5728\u7684\u6d4b\u8bd5\u6570\u636e\u6c61\u67d3\u95ee\u9898\uff1a\u516c\u5f00\u7684\u6d4b\u8bd5\u6570\u636e\uff0c\u5982 MMLU \u7b49\uff0c\u4f1a\u4e0d\u53ef\u907f\u514d\u5730\u88ab\u7eb3\u5165\u8bad\u7ec3\u96c6\u6216\u6709\u9488\u5bf9\u6027\u5730 overfit \u6d4b\u8bd5\u96c6\u3002\u201c\u5982\u4f55\u4fdd\u8bc1\u6d4b\u8bd5\u6570\u636e\u80fd\u591f\u5408\u7406\u4e14\u6b63\u786e\u7684\u8bc4\u4f30\u5927\u8bed\u8a00\u6a21\u578b\u201d\u5f15\u8d77\u4e86\u5b66\u672f\u754c\u7684\u5e7f\u6cdb\u5173\u6ce8\u3002<\/p>\n\n\n\n<p>\u5bf9\u6b64\uff0c\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86 DyVal\uff08Dynamic Evaluation\uff0c\u52a8\u6001\u8bc4\u6d4b\u534f\u8bae\uff09\uff0c\u8be5\u534f\u8bae\u5229\u7528\u6709\u5411\u65e0\u73af\u56fe\uff08directed acyclic graphs, DAGs\uff09\u52a8\u6001\u751f\u6210\u6d4b\u8bd5\u6570\u636e\uff0c\u4ece\u800c\u964d\u4f4e\u4e86\u6d4b\u8bd5\u6570\u636e\u88ab\u6a21\u578b\u8bb0\u5fc6\u7684\u53ef\u80fd\u6027\u3002\u6b64\u5916\uff0cDyVal \u751f\u6210\u7684\u8bc4\u6d4b\u6570\u636e\u8fd8\u53ef\u4ee5\u4f5c\u4e3a\u8bed\u8a00\u6a21\u578b\u7684\u6570\u636e\u589e\u5f3a\u624b\u6bb5\u3002\u4f8b\u5982\uff0c\u4f7f\u7528 DyVal \u751f\u6210\u7684\u6570\u636e\u5bf9 Llama2-7b \u6a21\u578b\u8fdb\u884c\u5fae\u8c03\uff0c\u53ef\u6709\u6548\u5730\u63d0\u5347\u6a21\u578b\u5728\u4f17\u591a\u63a8\u7406\u6570\u636e\u96c6\uff08\u5982GSM8K\u3001FOLIO\u7b49\uff09\u4e0a\u7684\u8868\u73b0\u3002\u52a8\u6001\u8bc4\u6d4b\u662f\u5927\u8bed\u8a00\u6a21\u578b\u8bc4\u6d4b\u7684\u4e00\u4e2a\u65b0\u65b9\u5411\uff0c\u7814\u7a76\u5458\u4eec\u671f\u5f85\u8d8a\u6765\u8d8a\u591a\u7684\u5de5\u4f5c\u51fa\u73b0\u5728\u8fd9\u4e00\u9886\u57df\uff0c\u4ee5\u5e2e\u52a9\u4eba\u4eec\u66f4\u597d\u5730\u7406\u89e3\u6a21\u578b\u7684\u5b9e\u9645\u80fd\u529b\u3002<\/p>\n\n\n\n<p>\u7814\u7a76\u7ed3\u679c\u8868\u660e\uff1a<\/p>\n\n\n\n<p>\u5927\u8bed\u8a00\u6a21\u578b\u5728\u73b0\u6709\u9759\u6001\u57fa\u51c6\u548c DyVal \u4e4b\u95f4\u7684\u6027\u80fd\u8868\u73b0\u5b58\u5728\u663e\u8457\u5dee\u5f02\uff1a\u4f8b\u5982\uff0cphi-1.5\u3001Xwin \u548c Wizard \u6a21\u578b\u5728\u73b0\u6709\u7684\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u53d6\u5f97\u4e86\u4f18\u5f02\u7684\u6210\u7ee9\uff0c\u4f46\u662f\u5b83\u4eec\u5728\u7814\u7a76\u5458\u4eec\u7684\u8bc4\u4f30\u4e2d\u8868\u73b0\u4e0d\u4f73\u3002\u8fd9\u4e00\u73b0\u8c61\u7a81\u663e\u4e86\u4ec5\u5728\u9759\u6001\u57fa\u51c6\u6d4b\u8bd5\u4e2d\u8bc4\u6d4b\u5927\u8bed\u8a00\u6a21\u578b\u65f6\u7684\u6f5c\u5728\u95ee\u9898\uff0c\u540c\u65f6\u4e5f\u63ed\u793a\u4e86\u53ef\u80fd\u5b58\u5728\u7684\u4f4e\u8bad\u7ec3\u6570\u636e\u8d28\u91cf\u6216\u6570\u636e\u6c61\u67d3\u7684\u98ce\u9669\u3002<\/p>\n\n\n\n<p>\u96be\u4ee5\u5e94\u4ed8\u590d\u6742\u6570\u636e\u96c6\uff1a\u4ece D1 \u5230 D4\uff0c\u6a21\u578b\u6027\u80fd\u663e\u8457\u4e0b\u964d\uff0c\u8fd9\u7a81\u663e\u4e86\u968f\u7740\u6570\u636e\u96c6\u590d\u6742\u5ea6\u7684\u63d0\u5347\uff0c\u5927\u8bed\u8a00\u6a21\u578b\u6240\u906d\u9047\u7684\u56f0\u96be\u8d8a\u6765\u8d8a\u5927\u3002\u4f8b\u5982\uff0c\u968f\u7740\u590d\u6742\u5ea6\u7684\u589e\u52a0\uff0cGPT-4 \u5728\u7b97\u672f\u4efb\u52a1\u4e0a\u7684\u6027\u80fd\u4e0b\u964d\u4e8623%\u3002\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0c\u6240\u6709\u6a21\u578b\u5728\u5f52\u7eb3\u903b\u8f91\uff08\u4ece\u7ed3\u8bba\u63a8\u5bfc\u524d\u63d0\uff09\u65b9\u9762\u7684\u8868\u73b0\u90fd\u666e\u904d\u4f4e\u4e8e\u6f14\u7ece\u903b\u8f91\uff08\u4ece\u524d\u63d0\u63a8\u5bfc\u7ed3\u8bba\uff09\uff0c\u8fd9\u4e00\u73b0\u8c61\u4e5f\u8bc1\u5b9e\u4e86\u5927\u6a21\u578b\u63a8\u5bfc &#8220;A -> B&#8221; \u65f6\u6bd4 &#8220;B -> A&#8221; \u65f6\u8868\u73b0\u66f4\u4e3a\u51fa\u8272\u3002\u6b64\u5916\uff0cGPT-4 \u548c GPT-3.5 \u4e4b\u95f4\u7684\u6027\u80fd\u5dee\u5f02\u867d\u7136\u5728\u50cf D1 \u8fd9\u6837\u7684\u7b80\u5355\u4efb\u52a1\u4e2d\u5fae\u4e0d\u8db3\u9053\uff0c\u4f46\u5728\u590d\u6742\u4efb\u52a1\u4e2d\u5374\u53d8\u5f97\u975e\u5e38\u660e\u663e\u3002\u8fd9\u8868\u660e\u6211\u4eec\u9700\u8981\u66f4\u590d\u6742\u7684\u591a\u4efb\u52a1\uff0c\u4ee5\u6709\u6548\u8bc4\u4f30\u6a21\u578b\u7684\u80fd\u529b\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"968\" height=\"525\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-6.png\" alt=\"DyVal: Dynamic Evaluation of Large Language Models for Reasoning Tasks | diagram\" class=\"wp-image-1075893\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-6.png 968w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-6-300x163.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-6-768x417.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-6-240x130.png 240w\" sizes=\"auto, (max-width: 968px) 100vw, 968px\" \/><figcaption class=\"wp-element-caption\">\u56fe3\uff1aDyVal \u793a\u610f\u56fe<\/figcaption><\/figure>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"kosmos-2-\u5c06\u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b\u540c\u89c6\u89c9\u4e16\u754c\u8fde\u63a5\u5bf9\u5e94\">KOSMOS-2\uff1a\u5c06\u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b\u540c\u89c6\u89c9\u4e16\u754c\u8fde\u63a5\u5bf9\u5e94<\/h3>\n\n\n\n<figure class=\"wp-block-image size-full is-resized\"><img loading=\"lazy\" decoding=\"async\" width=\"1304\" height=\"430\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-7.png\" alt=\"Kosmos-2: Grounding Multimodal Large Language Models to the World\n\" class=\"wp-image-1076751\" style=\"width:677px;height:auto\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-7.png 1304w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-7-300x99.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-7-1024x338.png 1024w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-7-768x253.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-7-240x79.png 240w\" sizes=\"auto, (max-width: 1304px) 100vw, 1304px\" \/><\/figure>\n\n\n\n<p>\u8bba\u6587\u94fe\u63a5\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/arxiv.org\/abs\/2306.14824\" target=\"_blank\" rel=\"noopener noreferrer\">https:\/\/arxiv.org\/abs\/2306.14824<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>Demo \u94fe\u63a5\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/build.nvidia.com\/microsoft\/microsoft-kosmos-2\" target=\"_blank\" rel=\"noopener noreferrer\">https:\/\/build.nvidia.com\/microsoft\/microsoft-kosmos-2<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>KOSMOS-2 \u662f\u4e00\u4e2a\u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b\uff0c\u5177\u5907\u4e24\u79cd\u65b0\u7684\u80fd\u529b\u2014\u2014Grounding \u548c Referring\u3002Grounding \u80fd\u529b\u53ef\u4ee5\u4f7f\u5f97\u6a21\u578b\u80fd\u591f\u901a\u8fc7\u4f8b\u5982\u8fb9\u754c\u6846\uff08bounding boxes\uff09\u7684\u65b9\u5f0f\uff0c\u5c06\u6587\u672c\u8f93\u51fa\u4e0e\u89c6\u89c9\u4e16\u754c\u4e2d\u7684\u7269\u4f53\u6216\u533a\u57df\u76f8\u8fde\u63a5\uff0c\u8fdb\u800c\u63d0\u4f9b\u66f4\u52a0\u4e30\u5bcc\u7684\u56de\u7b54\uff0c\u51cf\u5c11\u5171\u6307\u6b67\u4e49\uff0c\u5e76\u652f\u6301\u66f4\u591a\u7684\u89c6\u89c9-\u8bed\u8a00\u4efb\u52a1\u3002Referring \u5219\u5141\u8bb8\u7528\u6237\u901a\u8fc7\u4f8b\u5982\u8fb9\u754c\u6846\u7684\u65b9\u5f0f\uff0c\u9009\u62e9\u89c6\u89c9\u4e16\u754c\u4e2d\u7684\u7269\u4f53\u6216\u533a\u57df\u4f5c\u4e3a\u6a21\u578b\u7684\u8f93\u5165\uff0c\u4f46\u4e0d\u9700\u8981\u63d0\u4f9b\u8be6\u7ec6\u7684\u6587\u672c\u63cf\u8ff0\u6765\u6307\u4ee3\u5b83\u4eec\uff0c\u4ece\u800c\u5b9e\u73b0\u66f4\u52a0\u65b9\u4fbf\u7684\u4eba\u673a\u4ea4\u4e92\u3002\u4f9d\u6258\u4e8e Grounding \u548c Referring \u80fd\u529b\uff0cKOSMOS-2 \u63d0\u4f9b\u4e86\u4e00\u4e2a\u66f4\u7075\u6d3b\u3001\u66f4\u901a\u7528\u7684\u89c6\u89c9-\u8bed\u8a00\u4efb\u52a1\u4eba\u673a\u754c\u9762\u3002<\/p>\n\n\n\n<p>\u4e3a\u4e86\u89e3\u9501 KOSMOS-2 \u7684\u65b0\u80fd\u529b\uff0c\u7814\u7a76\u5458\u4eec\u57fa\u4e8e\u5927\u89c4\u6a21\u7684\u56fe\u50cf-\u6587\u672c\u5bf9\u6570\u636e\u6784\u5efa\u4e86 GRIT\uff08grounded image-text pairs\uff09 \u6570\u636e\u96c6\uff0c\u5c06\u6587\u672c\u63cf\u8ff0\u4e0e\u56fe\u7247\u4e2d\u7684\u7269\u4f53\u6216\u533a\u57df\u8fdb\u884c\u5bf9\u5e94\u8fde\u63a5\u3002\u7814\u7a76\u5458\u4eec\u628a\u7269\u4f53\u6216\u8005\u533a\u57df\u7684\u4f4d\u7f6e\u5750\u6807\u8f6c\u53d8\u6210\u4f4d\u7f6e\u6807\u8bb0\uff08location tokens\uff09\uff0c\u5e76\u901a\u8fc7\u201c\u8d85\u94fe\u63a5\u201d\u7684\u65b9\u5f0f\uff0c\u5c06\u6587\u672c\u63cf\u8ff0\u4e0e\u5bf9\u5e94\u7684\u4f4d\u7f6e\u6807\u8bb0\u8fde\u63a5\u5230\u4e00\u8d77\uff0c\u4f7f\u6a21\u578b\u80fd\u591f\u7406\u89e3\u5e76\u5b66\u4e60\u8fd9\u4e9b\u5bf9\u5e94\u5173\u7cfb\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"874\" height=\"520\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-8.png\" alt=\"KOSMOS-2: Grounding Multimodal Large Language Models to the World | diagram\" class=\"wp-image-1075845\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-8.png 874w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-8-300x178.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-8-768x457.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-8-240x143.png 240w\" sizes=\"auto, (max-width: 874px) 100vw, 874px\" \/><figcaption class=\"wp-element-caption\">\u56fe4\uff1aKOSMOS-2 \u6846\u67b6\u56fe<\/figcaption><\/figure>\n\n\n\n<p>KOSMOS-2 \u5728\u5f15\u5165\u65b0\u80fd\u529b\u7684\u540c\u65f6\uff0c\u4e5f\u4fdd\u7559\u4e86\u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b\u7684\u7684\u5e38\u89c4\u529f\u80fd\u3002\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0cKOSMOS-2 \u5728\u591a\u6a21\u6001 Grounding \u548c Referring \u4efb\u52a1\u4e0a\u53d6\u5f97\u4e86\u4f18\u5f02\u7684\u6210\u7ee9\uff0c\u540c\u65f6\u5728\u4e00\u4e9b\u57fa\u7840\u7684\u89c6\u89c9\u56fe\u50cf\u4efb\u52a1\u4ee5\u53ca\u81ea\u7136\u8bed\u8a00\u7406\u89e3\u548c\u751f\u6210\u65b9\u9762\u4e5f\u8868\u73b0\u51fa\u8272\u3002KOSMOS-2 \u878d\u5408\u4e86\u8bed\u8a00\u3001\u591a\u6a21\u6001\u611f\u77e5\u4e0e\u4e16\u754c\u5efa\u6a21\u7684\u80fd\u529b\uff0c\u6807\u5fd7\u7740\u8fc8\u5411\u4eba\u5de5\u901a\u7528\u667a\u80fd\u7684\u5173\u952e\u4e00\u6b65\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\" id=\"research5\">MG-TSD\uff1a\u57fa\u4e8e\u5f15\u5bfc\u5b66\u4e60\u8fc7\u7a0b\u7684\u591a\u7c92\u5ea6\u65f6\u95f4\u5e8f\u5217\u6269\u6563\u6a21\u578b<\/h3>\n\n\n\n<figure class=\"wp-block-image size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"856\" height=\"319\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-9.png\" alt=\"MG-TSD: Multi-Granularity Time Series Diffusion Models with Guided Learning Process\n\" class=\"wp-image-1076754\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-9.png 856w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-9-300x112.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-9-768x286.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/05\/new-arrival-in-research-11-9-240x89.png 240w\" sizes=\"auto, (max-width: 856px) 100vw, 856px\" \/><\/figure>\n\n\n\n<p>\u8bba\u6587\u94fe\u63a5\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/arxiv.org\/abs\/2403.05751\" target=\"_blank\" rel=\"noopener noreferrer\">https:\/\/arxiv.org\/abs\/2403.05751<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>\u9879\u76ee\u94fe\u63a5\uff1a<a class=\"msr-external-link glyph-append glyph-append-open-in-new-tab glyph-append-xsmall\" href=\"https:\/\/github.com\/Hundredl\/MG-TSD\" target=\"_blank\" rel=\"noopener noreferrer\">https:\/\/github.com\/Hundredl\/MG-TSD<span class=\"sr-only\"> (opens in new tab)<\/span><\/a><\/p>\n\n\n\n<p>\u65f6\u95f4\u5e8f\u5217\u9884\u6d4b\u5728\u91d1\u878d\u3001\u80fd\u6e90\u89c4\u5212\u3001\u6c14\u5019\u5efa\u6a21\u548c\u751f\u7269\u79d1\u5b66\u7b49\u591a\u4e2a\u9886\u57df\u90fd\u6709\u7740\u91cd\u8981\u5e94\u7528\u3002\u8fd1\u5e74\u6765\uff0c\u8bb8\u591a\u7814\u7a76\u5f00\u59cb\u91c7\u7528\u751f\u6210\u5f0f\u6a21\u578b\u6765\u89e3\u51b3\u65f6\u95f4\u5e8f\u5217\u9884\u6d4b\u95ee\u9898\uff0c\u5176\u4e2d\uff0c\u57fa\u4e8e\u6269\u6563\u6a21\u578b\u7684\u7814\u7a76\u56e0\u5176\u51fa\u8272\u7684\u6982\u7387\u9884\u6d4b\u6027\u8d28\u800c\u5907\u53d7\u5173\u6ce8\u3002\u7136\u800c\uff0c\u4e0e\u57fa\u4e8e\u81ea\u56de\u5f52\u6a21\u578b\u7684\u786e\u5b9a\u6027\u6a21\u578b\u76f8\u6bd4\uff0c\u6269\u6563\u6a21\u578b\u5728\u65f6\u95f4\u5e8f\u5217\u9884\u6d4b\u4efb\u52a1\u4e2d\u9762\u4e34\u7684\u4e00\u4e2a\u6311\u6218\u5728\u4e8e\uff0c\u5176\u968f\u673a\u6027\u5bfc\u81f4\u7684\u4e0d\u7a33\u5b9a\u6027\u66f4\u4e3a\u663e\u8457\u3002<\/p>\n\n\n\n<p>\u4e3a\u4e86\u89e3\u51b3\u6269\u6563\u6a21\u578b\u5728\u65f6\u95f4\u5e8f\u5217\u9884\u6d4b\u4e2d\u7684\u4e0d\u7a33\u5b9a\u6027\u95ee\u9898\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u65b0\u9896\u7684\u591a\u7c92\u5ea6\u65f6\u95f4\u5e8f\u5217\u6269\u6563\uff08MG-TSD\uff09\u6a21\u578b\u3002\u8be5\u6a21\u578b\u5229\u7528\u6570\u636e\u5185\u5728\u7684\u591a\u7c92\u5ea6\u6c34\u5e73\u4f5c\u4e3a\u4e2d\u95f4\u6269\u6563\u6b65\u9aa4\u7684\u76ee\u6807\uff0c\u4ee5\u5f15\u5bfc\u6269\u6563\u6a21\u578b\u7684\u5b66\u4e60\u8fc7\u7a0b\u3002\u7814\u7a76\u5458\u4eec\u6784\u5efa\u76ee\u6807\u7684\u65b9\u5f0f\u662f\u53d7\u5230\u4e86\u4e00\u4e2a\u89c2\u5bdf\u7684\u542f\u53d1\uff0c\u5373\u6269\u6563\u6a21\u578b\u7684\u524d\u5411\u8fc7\u7a0b\u9010\u6e10\u4f7f\u6570\u636e\u5206\u5e03\u9000\u5316\u5230\u6807\u51c6\u6b63\u6001\u5206\u5e03\uff0c\u8fd9\u4e00\u8fc7\u7a0b\u4e0e\u5c06\u7cbe\u7ec6\u6570\u636e\u5e73\u6ed1\u6210\u7c97\u7c92\u5ea6\u8868\u793a\u7684\u8fc7\u7a0b\u76f8\u543b\u5408\uff0c\u56e0\u4e3a\u8fd9\u4e24\u4e2a\u8fc7\u7a0b\u90fd\u5bfc\u81f4\u4e86\u7ec6\u5206\u5e03\u7279\u5f81\u7684\u9010\u6e10\u4e27\u5931\u3002<\/p>\n\n\n\n<p>\u5177\u4f53\u800c\u8a00\uff0c\u7814\u7a76\u5458\u4eec\u5f15\u5165\u4e86\u4e00\u4e2a\u65b0\u9896\u7684\u591a\u7c92\u5ea6\u5f15\u5bfc\u6269\u6563\u635f\u5931\u51fd\u6570\uff0c\u5e76\u63d0\u51fa\u4e86\u4e00\u79cd\u7b80\u6d01\u7684\u5b9e\u73b0\u65b9\u6cd5\uff0c\u6765\u6709\u6548\u5229\u7528\u4e0d\u540c\u7c92\u5ea6\u6c34\u5e73\u4e0a\u7684\u7c97\u7c92\u5ea6\u6570\u636e\u3002\u7814\u7a76\u5458\u4eec\u8bbe\u5b9a\u4e86\u81ea\u76d1\u7763\u7684\u5b66\u4e60\u76ee\u6807\u4f5c\u4e3a\u4e2d\u95f4\u6f5c\u5728\u72b6\u6001\u7684\u7ea6\u675f\uff0c\u5e76\u4f7f\u5176\u5f62\u6210\u4e00\u4e2a\u6b63\u5219\u5316\u7684\u91c7\u6837\u8def\u5f84\uff0c\u4ece\u800c\u4fdd\u7559\u4e86\u7c97\u7c92\u5ea6\u6570\u636e\u5185\u7684\u8d8b\u52bf\u548c\u6a21\u5f0f\u3002\u901a\u8fc7\u5f15\u5165\u8fd9\u79cd\u5f52\u7eb3\u504f\u5dee\uff0c\u7814\u7a76\u5458\u4eec\u4fc3\u8fdb\u4e86\u5728\u4e2d\u95f4\u6b65\u9aa4\u4e2d\u751f\u6210\u66f4\u7c97\u7684\u7279\u5f81\uff0c\u5e76\u6709\u52a9\u4e8e\u5728\u968f\u540e\u7684\u6269\u6563\u6b65\u9aa4\u4e2d\u6062\u590d\u66f4\u7ec6\u7684\u7279\u5f81\u3002\u56e0\u6b64\uff0c\u8fd9\u79cd\u8bbe\u8ba1\u964d\u4f4e\u4e86\u4e0d\u7a33\u5b9a\u6027\uff0c\u4ea7\u751f\u4e86\u9ad8\u8d28\u91cf\u7684\u9884\u6d4b\u7ed3\u679c\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image aligncenter size-full\"><img loading=\"lazy\" decoding=\"async\" width=\"777\" height=\"384\" src=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-10.png\" alt=\"MG-TSD: Multi-granularity Time Series Diffusion Models with Guided Learning Process | diagram\" class=\"wp-image-1075851\" srcset=\"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-10.png 777w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-10-300x148.png 300w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-10-768x380.png 768w, https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-content\/uploads\/2024\/08\/ICLR_new-arrival-in-research-11-10-240x119.png 240w\" sizes=\"auto, (max-width: 777px) 100vw, 777px\" \/><figcaption class=\"wp-element-caption\">\u56fe5\uff1a\u591a\u7c92\u5ea6\u65f6\u95f4\u5e8f\u5217\u6269\u6563\uff08MG-TSD\uff09\u6a21\u578b\u6846\u67b6\uff0c\u5305\u62ec\u4e09\u4e2a\u5173\u952e\u6a21\u5757\uff1a\u591a\u7c92\u5ea6\u6570\u636e\u751f\u6210\u5668\u3001\u65f6\u95f4\u8fc7\u7a0b\u6a21\u5757\uff08TPM\uff09\u548c\u7528\u4e8e\u7279\u5b9a\u7c92\u5ea6\u7ea7\u522b\u65f6\u95f4\u5e8f\u5217\u9884\u6d4b\u7684\u5f15\u5bfc\u6269\u6563\u8fc7\u7a0b\u6a21\u5757<\/figcaption><\/figure>\n\n\n\n<p>\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0cMG-TSD \u6a21\u578b\u5728\u4e0d\u540c\u9886\u57df\u7684\u771f\u5b9e\u4e16\u754c\u7684\u65f6\u95f4\u5e8f\u5217\u9884\u6d4b\u4efb\u52a1\u4e0a\u90fd\u80fd\u591f\u6709\u6548\u5730\u8fdb\u884c\u9884\u6d4b\uff0c\u5e76\u5177\u6709\u4f18\u5f02\u7684\u7a33\u5b9a\u6027\u548c\u9c81\u68d2\u6027\u3002\u6b64\u5916\uff0c\u8be5\u6a21\u578b\u4e0d\u4f9d\u8d56\u4e8e\u989d\u5916\u7684\u5916\u90e8\u6570\u636e\uff0c\u4f7f\u5176\u5728\u5404\u79cd\u9886\u57df\u90fd\u5177\u6709\u5e7f\u6cdb\u7684\u9002\u7528\u6027\u3002<\/p>\n\n\n\n<p>MG-TSD \u6a21\u578b\u4e3a\u89e3\u51b3\u65f6\u95f4\u5e8f\u5217\u9884\u6d4b\u4e2d\u7684\u4e0d\u7a33\u5b9a\u6027\u95ee\u9898\u63d0\u4f9b\u4e86\u4e00\u4e2a\u521b\u65b0\u7684\u89e3\u51b3\u65b9\u6848\uff0c\u4e3a\u672a\u6765\u7684\u7814\u7a76\u548c\u5e94\u7528\u63d0\u4f9b\u4e86\u65b0\u7684\u65b9\u5411\u548c\u53ef\u80fd\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u7f16\u8005\u6309\uff1a\u6b22\u8fce\u9605\u8bfb\u201c\u79d1\u7814\u4e0a\u65b0\u201d\u680f\u76ee\uff01\u201c\u79d1\u7814\u4e0a\u65b0\u201d\u6c47\u805a\u4e86\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u6700\u65b0\u7684\u521b\u65b0\u6210\u679c\u4e0e\u79d1\u7814\u52a8\u6001\u3002\u5728\u8fd9\u91cc\uff0c\u4f60\u53ef\u4ee5\u5feb\u901f\u6d4f\u89c8\u7814\u7a76\u9662\u7684\u4eae\u70b9\u8d44\u8baf\uff0c\u4fdd\u6301\u5bf9\u524d\u6cbf\u9886\u57df\u7684\u654f\u9510\u55c5\u89c9\uff0c\u540c\u65f6\u4e5f\u80fd\u627e\u5230\u5148\u8fdb\u5b9e\u7528\u7684\u5f00\u6e90\u5de5\u5177\u3002 \u672c\u5468\uff0c\u5168\u7403\u6700\u8d1f\u76db\u540d\u7684\u4eba\u5de5\u667a\u80fd\u76db\u4f1a\u4e4b\u4e00 ICLR \u5927\u4f1a\u5c06\u5728\u5965\u5730\u5229\u7ef4\u4e5f\u7eb3\u4e3e\u529e\u3002\u6240\u4ee5\uff0c\u4eca\u5929\u7684\u201c\u79d1\u7814\u4e0a\u65b0\u201d\u5c06\u4e3a\u5927\u5bb6\u5e26\u6765\u591a\u7bc7\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u5728 ICLR 2024 \u4e0a\u7684\u7cbe\u9009\u8bba\u6587\u89e3\u8bfb\uff0c\u6d89\u53ca\u9886\u57df\u6db5\u76d6\u6df1\u5ea6\u5f3a\u5316\u5b66\u4e60\u3001\u591a\u6a21\u6001\u8bed\u8a00\u6a21\u578b\u3001\u65f6\u95f4\u5e8f\u5217\u6269\u6563\u6a21\u578b\u3001\u65e0\u76d1\u7763\u5b66\u4e60\u7b49\u591a\u4e2a\u524d\u6cbf\u4e3b\u9898\u3002 \u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/openreview.net\/forum?id=Z8UfDs4J46 (opens in new tab) \u8fd1\u5e74\u6765\uff0c\u6df1\u5ea6\u5f3a\u5316\u5b66\u4e60\uff08DRL\uff09\u53ca\u5176\u5e94\u7528\u8fc5\u901f\u53d1\u5c55\uff0c\u5b83\u4e0d\u4ec5\u5728\u865a\u62df\u4efb\u52a1\uff08\u5982\u89c6\u9891\u6e38\u620f\u548c\u6a21\u62df\u673a\u5668\u4eba\u73af\u5883\uff09\u4e0a\u53d6\u5f97\u4e86\u6210\u529f\uff0c\u4e5f\u5728\u8bb8\u591a\u5177\u6709\u6311\u6218\u6027\u7684\u73b0\u5b9e\u4e16\u754c\u4efb\u52a1\u4e2d\u5f97\u5230\u4e86\u8bc1\u660e\uff0c\u4f8b\u5982\u63a7\u5236\u6258\u5361\u9a6c\u514b\u548c\u901a\u8fc7\u4eba\u7c7b\u53cd\u9988\u8c03\u6574\u5927\u8bed\u8a00\u6a21\u578b\u3002\u7136\u800c\uff0c\u5bfc\u81f4\u667a\u80fd\u4f53\u53ef\u80fd\u65e0\u6cd5\u7acb\u5373\u89c2\u5bdf\u5230\u5f53\u524d\u73af\u5883\u72b6\u6001\u6216\u5176\u884c\u52a8\u65e0\u6cd5\u7acb\u5373\u5f71\u54cd\u73af\u5883\u7684\u4fe1\u53f7\u5ef6\u8fdf\uff0c\u5728\u6df1\u5ea6\u5f3a\u5316\u5b66\u4e60\u7814\u7a76\u4e2d\u957f\u671f\u5b58\u5728\u4e14\u7ecf\u5e38\u88ab\u5ffd\u89c6\u3002\u8be5\u95ee\u9898\u5e7f\u6cdb\u5b58\u5728\u4e8e\u5404\u79cd\u5b9e\u9645\u5e94\u7528\u4e2d\uff0c\u5bf9\u57fa\u4e8e\u6df1\u5ea6\u5f3a\u5316\u5b66\u4e60\u89e3\u51b3\u65b9\u6848\u7684\u6709\u6548\u6027\u4ea7\u751f\u4e86\u91cd\u5927\u5f71\u54cd\uff0c\u56e0\u6b64\u8be5\u6311\u6218\u8feb\u5207\u9700\u8981\u7814\u7a76\u8fdb\u884c\u5e94\u5bf9\u3002 \u4e3a\u4e86\u89e3\u51b3 DRL \u4e2d\u7684\u4fe1\u53f7\u5ef6\u8fdf\u95ee\u9898\uff0c\u7814\u7a76\u5458\u4eec\u9996\u5148\u901a\u8fc7\u6269\u5c55\u9a6c\u5c14\u53ef\u592b\u51b3\u7b56\u8fc7\u7a0b\u6846\u67b6\u6765\u5b9a\u4e49\u5ef6\u8fdf\u89c2\u6d4b\u9a6c\u5c14\u53ef\u592b\u51b3\u7b56\u8fc7\u7a0b\uff08DOMDP\uff09\uff0c\u4ece\u800c\u5c06\u4fe1\u53f7\u5ef6\u8fdf\u7684\u60c5\u51b5\u7eb3\u5165\u8003\u8651\u4e4b\u4e2d\u3002\u7136\u540e\uff0c\u7814\u7a76\u5458\u4eec\u5728\u8bba\u6587\u4e2d\u9610\u660e\u4e86 DRL \u91cc\u4fe1\u53f7\u5ef6\u8fdf\u5b58\u5728\u7684\u6311\u6218\uff0c\u5e76\u5c55\u793a\u4e86\u5e38\u89c4 DRL \u7b97\u6cd5\u548c\u90e8\u5206\u53ef\u89c2\u6d4b\u9a6c\u5c14\u53ef\u592b\u51b3\u7b56\u8fc7\u7a0b\uff08POMDP\uff09\u7684\u901a\u7528\u65b9\u6cd5\u53d7\u5230\u5ef6\u8fdf\u7684\u4e25\u91cd\u5f71\u54cd\u3002 \u9488\u5bf9\u8fd9\u4e9b\u6311\u6218\uff0c\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u4e00\u7cfb\u5217\u65b0\u65b9\u6cd5\uff0c\u65e8\u5728\u63d0\u9ad8\u5b58\u5728\u5ef6\u8fdf\u65f6 DRL \u7b97\u6cd5\u7684\u6027\u80fd\u3002\u7ed3\u5408\u7406\u8bba\u89c1\u89e3\u548c\u5b9e\u9645\u7b97\u6cd5\u8c03\u6574\uff0c\u7814\u7a76\u5458\u4eec\u6269\u5c55\u4e86\u4f20\u7edf\u7684 actor-critic \u6846\u67b6\uff0c\u5e76\u63d0\u51fa\u4e86\u6709\u6548\u7684\u7b56\u7565\u6765\u514b\u670d\u8fd9\u4e9b\u6311\u6218\u3002\u5145\u5206\u7684\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u5728\u5177\u6709\u8f83\u5927\u5ef6\u8fdf\u7684\u8fde\u7eed\u673a\u5668\u4eba\u63a7\u5236\u4efb\u52a1\u4e2d\uff0c\u91c7\u7528\u8be5\u8bba\u6587\u63d0\u51fa\u7684\u65b9\u6cd5\u540e\uff0cDRL \u7b97\u6cd5\u53d6\u5f97\u4e86\u5353\u8d8a\u7684\u6027\u80fd\uff0c\u5176\u7ed3\u679c\u4e0e\u65e0\u5ef6\u8fdf\u60c5\u51b5\u76f8\u6bd4\uff0c\u6027\u80fd\u635f\u5931\u8f83\u5c0f\u3002 \u8fd9\u9879\u7814\u7a76\u5728\u89e3\u51b3 DRL \u4e2d\u4e00\u4e2a\u57fa\u672c\u6311\u6218\u65b9\u9762\u8fc8\u51fa\u4e86\u91cd\u8981\u7684\u4e00\u6b65\uff0c\u4e0d\u4ec5\u62d3\u5bbd\u4e86\u5176\u5728\u73b0\u5b9e\u73af\u5883\u4e2d\u7684\u5e94\u7528\u8303\u56f4\uff0c\u4e5f\u4e3a\u81ea\u4e3b\u7cfb\u7edf\u7684\u6301\u7eed\u53d1\u5c55\u505a\u51fa\u4e86\u8d21\u732e\u3002\u901a\u8fc7\u5f00\u53d1\u6709\u6548\u5e94\u5bf9\u4fe1\u53f7\u5ef6\u8fdf\u7684\u65b9\u6cd5\uff0c\u7814\u7a76\u5458\u4eec\u589e\u5f3a\u4e86 DRL \u7684\u5b9e\u7528\u6027\u548c\u53ef\u9760\u6027\uff0c\u4e3a\u5176\u5728\u975e\u7406\u60f3\u6761\u4ef6\u4e0b\u7684\u5e94\u7528\u5960\u5b9a\u4e86\u57fa\u7840\u3002 \u8bba\u6587\u94fe\u63a5\uff1ahttps:\/\/arxiv.org\/abs\/2401.08961 (opens in new tab) \u8fd1\u5e74\u6765\uff0c\u4e00\u79cd\u540d\u4e3a\u7ea7\u8054\u591a\u81c2\u8001\u864e\u673a\uff08cascading bandits\uff09[Kveton et al., 2015]\u7684\u6a21\u578b\u53d7\u5230\u4e86\u5e7f\u6cdb\u5173\u6ce8\uff0c\u5728\u63a8\u8350\u7cfb\u7edf\u3001\u5728\u7ebf\u5e7f\u544a\u4e2d\u5e94\u7528\u5e7f\u6cdb\u3002\u5728\u7ea7\u8054\u591a\u81c2\u8001\u864e\u673a\u4e2d\uff0c\u667a\u80fd\u4f53\u9700\u8981\u5728\u4f17\u591a\u9009\u9879\u4e2d\u6311\u9009\u4e00\u4e2a\u9009\u9879\u5217\u8868\u63a8\u8350\u7ed9\u7528\u6237\uff0c\u6bcf\u4e2a\u9009\u9879\u90fd\u6709\u4e00\u4e2a\u672a\u77e5\u7684\u5438\u5f15\u6982\u7387\u3002\u667a\u80fd\u4f53\u7684\u76ee\u6807\u5219\u662f\u4e0d\u65ad\u4f18\u5316\u63a8\u8350\u7684\u9009\u9879\u5217\u8868\uff0c\u4ee5\u6700\u5927\u5316\u671f\u671b\u7d2f\u79ef\u5956\u52b1\uff08\u70b9\u51fb\u7387\uff09\u3002 \u7136\u800c\uff0c\u73b0\u6709\u7684\u7ea7\u8054\u591a\u81c2\u8001\u864e\u673a\u6a21\u578b\u5ffd\u7565\u4e86\u7528\u6237\u72b6\u6001\uff08\u5982\u7528\u6237\u5386\u53f2\u884c\u4e3a\uff09\u5bf9\u63a8\u8350\u7684\u5f71\u54cd\uff0c\u4ee5\u53ca\u7528\u6237\u72b6\u6001\u53ef\u80fd\u7684\u6539\u53d8\u3002\u4e3a\u4e86\u89e3\u51b3\u8fd9\u4e00\u95ee\u9898\uff0c\u5fae\u8f6f\u4e9a\u6d32\u7814\u7a76\u9662\u7684\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u4e00\u79cd\u540d\u4e3a\u7ea7\u8054\u5f3a\u5316\u5b66\u4e60\uff08cascading reinforcement learning\uff09\u7684\u6a21\u578b\u3002\u5728\u8be5\u6a21\u578b\u4e2d\uff0c\u6bcf\u4e2a\u7528\u6237\u72b6\u6001-\u9009\u9879\u7684\u5339\u914d\u5bf9\u6709\u4e00\u4e2a\u672a\u77e5\u7684\u5438\u5f15\u6982\u7387\u3001\u4e00\u4e2a\u672a\u77e5\u7684\u72b6\u6001\u8f6c\u79fb\u5206\u5e03\u548c\u4e00\u4e2a\u5956\u52b1\u3002\u5982\u56fe2\u6240\u793a\uff0c\u5728\u6bcf\u4e2a\u65f6\u523b\uff0c\u667a\u80fd\u4f53\u4f1a\u5148\u89c2\u5bdf\u5230\u5f53\u524d\u7684\u7528\u6237\u72b6\u6001\uff0c\u7136\u540e\u63a8\u8350\u4e00\u4e2a\u957f\u5ea6\u4e3a m \u7684\u9009\u9879\u5217\u8868\u3002\u5982\u679c\u7528\u6237\u5bf9\u67d0\u4e00\u9009\u9879\u611f\u5174\u8da3\u5e76\u70b9\u51fb\uff0c\u90a3\u4e48\u7528\u6237\u5c06\u8f6c\u79fb\u5230\u4e0b\u4e00\u72b6\u6001\uff0c\u667a\u80fd\u4f53\u5219\u4f1a\u83b7\u5f97\u4e00\u4e2a\u5956\u52b1\u3002\u667a\u80fd\u4f53\u7684\u76ee\u6807\u662f\u6700\u5927\u5316\u671f\u671b\u7d2f\u79ef\u5956\u52b1\uff0c\u56e0\u6b64\u8be5\u6a21\u578b\u80fd\u6709\u6548\u5730\u5c06\u7528\u6237\u72b6\u6001\u53ca\u5176\u53d8\u5316\u7eb3\u5165\u63a8\u8350\u8fc7\u7a0b\u4e2d\u3002 \u9488\u5bf9\u8be5\u6a21\u578b\uff0c\u7814\u7a76\u5458\u4eec\u9996\u5148\u57fa\u4e8e\u52a8\u6001\u89c4\u5212\u8bbe\u8ba1\u4e86\u4e00\u4e2a\u5feb\u901f\u79bb\u7ebf\u6c42\u89e3\u5668 BestPerm\uff0c\u5b83\u80fd\u591f\u5728\u591a\u9879\u5f0f\u65f6\u95f4\u5185\u8ba1\u7b97\u51fa\u6700\u4f18\u7684\u9009\u9879\u5217\u8868\u3002\u7136\u540e\uff0c\u7814\u7a76\u5458\u4eec\u63d0\u51fa\u4e86\u5f3a\u5316\u5b66\u4e60\u7b97\u6cd5 CascadingVI\uff0c\u8be5\u7b97\u6cd5\u80fd\u591f\u8fbe\u5230 O \u0303(H\u221aHSNK) \u7684\u540e\u6094\u5ea6\uff08regret\uff09\u4e0a\u754c\uff0c\u8fd9\u4e2a\u7ed3\u679c\u53ea\u4f9d\u8d56\u4e8e\u9009\u9879\u7684\u4e2a\u6570 \ud835\udc41, \u800c\u4e0e\u9009\u9879\u5217\u8868\u7684\u4e2a\u6570\uff08\u7ea6 N^m\uff09\u65e0\u5173\u3002\u56e0\u6b64\uff0c\u8be5\u7b97\u6cd5\u80fd\u540c\u65f6\u4fdd\u8bc1\u91c7\u6837\u548c\u8ba1\u7b97\u7684\u9ad8\u6548\u6027\u3002 [&hellip;]<\/p>\n","protected":false},"author":42735,"featured_media":1076739,"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":0,"footnotes":""},"research-area":[],"msr-locale":[268881],"msr-post-option":[],"class_list":["post-1075818","msr-blog-post","type-msr-blog-post","status-publish","has-post-thumbnail","hentry","msr-locale-zh_cn"],"msr_assoc_parent":{"id":1012650,"type":"lab"},"_links":{"self":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1075818","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post"}],"about":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-blog-post"}],"author":[{"embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/users\/42735"}],"version-history":[{"count":4,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1075818\/revisions"}],"predecessor-version":[{"id":1076760,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-blog-post\/1075818\/revisions\/1076760"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/media\/1076739"}],"wp:attachment":[{"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1075818"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1075818"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1075818"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/newed.any0.dpdns.org\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1075818"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}