{"id":63049,"date":"2022-12-23T18:09:43","date_gmt":"2022-12-23T09:09:43","guid":{"rendered":"https:\/\/smilegate.ai\/?p=63049"},"modified":"2022-12-23T18:22:15","modified_gmt":"2022-12-23T09:22:15","slug":"%ea%b2%8c%ec%9e%84-%ec%98%81%ec%83%81%eb%a7%8c-%ea%b0%80%ec%a7%80%ea%b3%a0-%eb%a7%88%ec%9d%b8%ed%81%ac%eb%9e%98%ed%94%84%ed%8a%b8-%ed%95%99%ec%8a%b5-%ed%95%98%ea%b8%b0-video-pretraining-by-openai","status":"publish","type":"post","link":"https:\/\/smilegate.ai\/cn\/2022\/12\/23\/%ea%b2%8c%ec%9e%84-%ec%98%81%ec%83%81%eb%a7%8c-%ea%b0%80%ec%a7%80%ea%b3%a0-%eb%a7%88%ec%9d%b8%ed%81%ac%eb%9e%98%ed%94%84%ed%8a%b8-%ed%95%99%ec%8a%b5-%ed%95%98%ea%b8%b0-video-pretraining-by-openai\/","title":{"rendered":"\uac8c\uc784 \uc601\uc0c1\ub9cc \uac00\uc9c0\uace0 \ub9c8\uc778\ud06c\ub798\ud504\ud2b8 \ud559\uc2b5 \ud558\uae30 (Video PreTraining by OpenAI)"},"content":{"rendered":"

[\uac00\uc0c1\uc0dd\uba85\uc5f0\uad6c\ud300 \ubc15\uc8fc\ud615]<\/p>\n\n\n\n

\ucd5c\uadfc \uac15\ud654\ud559\uc2b5\uc5d0\uc11c \ub9ce\uc774 \uc8fc\ubaa9 \ubc1b\uace0 \uc788\ub294 \ubd84\uc57c \uc911\uc5d0 \ud558\ub098\uac00 Offline \ud559\uc2b5 \uc785\ub2c8\ub2e4. \uae30\uc874 \uac15\ud654\ud559\uc2b5\uc740 \uc2dc\ubbac\ub808\uc774\ud130\ub97c \ud1b5\ud574 \uc2e4\uc2dc\uac04\uc73c\ub85c trajectory\ub97c \ubc1b\uc544\uc11c \ud559\uc2b5\uc744 \ud588\uc5c8\uc5b4\uc57c \ud588\ub294\ub370\uc694, \uc2dc\ubbac\ub808\uc774\ud130 \ub9cc\ub4e4\uae30\ub3c4 \uc27d\uc9c0 \uc54a\uace0 \uc2e4\uc81c \uc801\uc6a9\ub3c4 \uc5b4\ub824\uc6cc\uc11c \ub85c\ubd07\uc774\ub098 \uc790\uc728 \uc8fc\ud589 \ucabd\uc5d0\uc11c\ub294 \ub9ce\uc774 \uc2dc\ub3c4 \ub418\ub294 \ubc29\ubc95 \uc785\ub2c8\ub2e4. \uc774\ub97c \uc704\ud574 \uac8c\uc784 \ucabd\uc5d0\uc11c\ub294 \ub2e4\uc591\ud55c \uc720\uc800\ub4e4\uc774 \ub9cc\ub4e4\uc5b4 \ub0b4\ub294 \ub85c\uadf8\ub97c \ud1b5\ud574\uc11c offline reinforcement learning\uc774\ub098 imitation learning\uc744 \uc2dc\ub3c4\ud558\uae30\ub3c4 \ud558\ub294 \ub370\uc694, \uc720\uc800 \ub85c\uadf8\ub97c \uc77c\ubc18\uc778\uc774 \uad6c\ud558\uae30\uac00 \uc27d\uc9c0\ub3c4 \uc54a\uace0 \ub85c\uadf8\uac00 \ub9ce\uc9c0 \uc54a\uc73c\uba74 \ud559\uc2b5\ub3c4 \uc5b4\ub824\uc6cc\uc11c \uc5ec\ub7ec\ubaa8\ub85c \uc9c4\uc785 \uc7a5\ubcbd\uc774 \uc788\ub294 \ubc29\ubc95 \uc774\uc5c8\uc2b5\ub2c8\ub2e4. <\/p>\n\n\n\n

\uc774 \ub54c \uac8c\uc784\uc73c\ub85c \uac15\ud654\ud559\uc2b5\uc744 \ud574 \ubcf4\uc2e0 \ubd84\uc774\ub77c\uba74 \ud55c\ubc88\ucbe4\uc740 \ub5a0\uc62c\ub824 \ubcf4\uc168\uc744 \ub4ef\ud55c \uc544\uc774\ub514\uc5b4\uac00 \uc788\uc2b5\ub2c8\ub2e4. \uc778\ud130\ub137\uc5d0 \uac8c\uc784\uc601\uc0c1\uc774 \uadf8\ub807\uac8c\ub098 \ub9ce\uc740\ub370 \ud574\ub2f9 \uc601\uc0c1\uc744 \uac00\uc9c0\uace0 \uac15\ud654\ud559\uc2b5\uc744 \ud574 \ubcfc\uc218\ub294 \uc5c6\uc744\uae4c? \uc0ac\uc2e4 \uac15\ud654\ud559\uc2b5\uc744 \ud558\uae30 \uc704\ud574\uc11c\ub294 observation \ubfd0\ub9cc \uc544\ub2c8\ub77c action\uc774 \ud544\uc694 \ud55c\ub370\uc694, \uc601\uc0c1\uc73c\ub85c\ub9cc action\uc744 \uacc4\uc0b0\ud55c\ub2e4\ub294\uac8c \ub9e4\uc6b0 \ud798\ub4e4\uae30 \ub54c\ubb38\uc5d0 \uc2e4\uc81c\ub85c\ub294 \uc801\uc6a9\ud558\uae30 \ud798\ub4e0 \uc544\uc774\ub514\uc5b4 \uc911\uc5d0 \ud558\ub098\uc600\uc2b5\ub2c8\ub2e4. \ud558\uc9c0\ub9cc OpenAI\uc5d0\uc11c\ub294 Video PreTraining (VPT)\ub77c\ub294 \ubc29\ubc95\uc744 \uace0\uc548\ud574 \uc774 \uc544\uc774\ub514\uc5b4\uac00 \uc2e4\ud604 \uac00\ub2a5\ud55c \uc544\uc774\ub514\uc5b4\ub77c\ub294 \uac83\uc744 \ubcf4\uc5ec \uc8fc\uc5c8\uc2b5\ub2c8\ub2e4. <\/p>\n\n\n\n

<\/div>\n\n\n\n

Learning to Play Minecraft with Video PreTraining (VPT)<\/h3>\n\n\n\n

OpenAI\ub294 \uc778\ud130\ub137 \uc601\uc0c1\uc744 \ud559\uc2b5\uc5d0 \uc774\uc6a9\ud558\uae30 \uc704\ud574 Video PreTraining (VPT) \ub77c\ub294 semi-supervised imitation learning \ubc29\ubc95\uc744 \uace0\uc548 \ud558\uc600\uc2b5\ub2c8\ub2e4. \uae30\ubcf8\uc801\uc778 \uc544\uc774\ub514\uc5b4\ub294 \ub9e4\uc6b0 \uac04\ub2e8 \ud569\ub2c8\ub2e4. \uc0ac\ub78c\ub4e4\uc5d0\uac8c \uc77c\uc815 \uc2dc\uac04\ub3d9\uc548 \ub9c8\uc778\ud06c\ub798\ud504\ud2b8 \uac8c\uc784\uc744 \uc2dc\ucf1c observation \ubfd0\ub9cc \uc544\ub2c8\ub77c action \uac12\uc744 \uc5bb\uc2b5\ub2c8\ub2e4. \uc774 \ub370\uc774\ud130\ub85c \uc774\ubbf8\uc9c0\ub9cc \ub4e4\uc5b4\uc624\uba74 \uc0ac\uc6a9\ub41c action\uc744 \uc608\uce21\ud560 \uc218 \uc788\ub294 Inverse Dynamis Model (IDM)\uc744 \ud559\uc2b5 \uc2dc\ud0b5\ub2c8\ub2e4. \uadf8\ub807\uac8c \ub41c\ub2e4\uba74 \uc778\ud130\ub137 \ub3d9\uc601\uc0c1\uc5d0\uc11c action\uc744 \ucda9\ubd84\ud788 \uc720\ucd94 \uac00\ub2a5\ud558\uae30 \ub54c\ubb38\uc5d0 \uc778\ud130\ub137 \ub3d9\uc601\uc0c1\uc744 behavior cloning \ub4f1\uc5d0 \uc5bc\ub9c8\ub4e0\uc9c0 \ud65c\uc6a9 \ud560 \uc218 \uc788\uac8c \ub429\ub2c8\ub2e4. <\/p>\n\n\n\n

\uc5ec\uae30\uc11c \ub610 \ud558\ub098\uc758 \uc911\uc694\ud55c \uc810\uc740 IDM\uc774 action\uc744 \uc608\uce21 \ud558\uae30 \uc704\ud574 \uacfc\uac70 frame\ub9cc \uc4f4\uac8c \uc544\ub2c8\ub77c \ubbf8\ub798 frame\ub3c4 \uac19\uc774 \uc0ac\uc6a9 \ud558\uc600\ub2e4\ub294 \uc810\uc785\ub2c8\ub2e4. \uc774\ub294 non-causal \ud615\ud0dc\uc758 \ud559\uc2b5\uc774 \uac00\ub2a5\ud558\uac8c \ud558\uc600\ub294\ub370\uc694, \uc774\ub85c \uc778\ud574 \ub2e8\uc21c\ud788 \uacfc\uac70\uc758 frame\ub9cc\uc73c\ub85c \ud604\uc7ac\uc758 action\uc744 \uc608\uc0c1 \ud558\ub294 BC \uac19\uc740 \ubc29\uc2dd \ubcf4\ub2e4 \ud6e8\uc52c \uc801\uc740 \uc591\uc758 \ub370\uc774\ud130\ub85c\ub3c4 \uc608\uce21\uc774 \uac00\ub2a5\ud558\ub3c4\ub85d \ud559\uc2b5 \ud560 \uc218 \uc788\uc5c8\uc2b5\ub2c8\ub2e4. <\/p>\n\n\n

\n
\"\"
VPT Method Overview (\ucd9c\ucc98: https:\/\/openai.com\/blog\/vpt\/<\/a>, OpenAI \uacf5\uc2dd \ube14\ub85c\uadf8)<\/figcaption><\/figure><\/div>\n\n\n

70,000 \uc2dc\uac04\uc758 \uc601\uc0c1\uc73c\ub85c \ud559\uc2b5\ub41c VPT Foundation Model \uc790\uccb4\ub9cc\uc73c\ub85c\ub3c4 \uac15\ud654\ud559\uc2b5\uc73c\ub85c\ub294 \uc774\ub8e8\uc5b4 \ub0bc \uc218 \uc5c6\ub294 \uc5c4\uccad\ub09c \uc131\ub2a5\uc744 \ubcf4\uc5ec \uc92c\uc2b5\ub2c8\ub2e4. \ub098\ubb34\ub97c \uc798\ub77c\uc11c \ud1b5\ub098\ubb34\ub97c \ubaa8\uc73c\ub294 \uac83 \ubfd0\ub9cc \uc544\ub2c8\ub77c, \uc774\ub97c \uc0ac\uc6a9\ud558\uc5ec \ud310\uc790\ub3c4 \ub9cc\ub4e4\uace0 \uc2ec\uc9c0\uc5b4\ub294 \ud14c\uc774\ube14\ub3c4 \ub9cc\ub4e4\uc5c8\uc2b5\ub2c8\ub2e4. \uc77c\ubc18\uc801\uc778 \uac15\ud654\ud559\uc2b5\uc5d0\uc11c\ub294 \uc774\ub807\uac8c \ud559\uc2b5\ub418\uae30\uac00 \uac70\uc758 \ubd88\uac00\ub2a5 \ud55c\ub370\uc694, \ud14c\uc774\ube14\uc744 \ub9cc\ub4e4\uae30 \uc704\ud574\uc11c\ub294 \uc57d 1000\uac1c\uc758 action\uc744 \uc5f0\uc18d\uc73c\ub85c \ud574\uc57c \ud558\ub294\ub370, \ud655\ub960\uc801\uc73c\ub85c \uc774\ub7ec\ud55c \uc77c\uc774 \uc77c\uc5b4\ub098\uae30\uac00 \uac70\uc758 \ubd88\uac00\ub2a5 \ud558\uae30 \ub54c\ubb38\uc785\ub2c8\ub2e4. <\/p>\n\n\n\n

\ucd94\uac00\ub85c \uc800\uc790\ub294 BC\uc758 \uc131\ub2a5\uc744 \uace0\ub3c4\ud654 \ud558\uae30 \uc704\ud574 fine-tuning\ub3c4 \ud14c\uc2a4\ud2b8 \ud574 \ubcf4\uc558\uc2b5\ub2c8\ub2e4. \uc0ac\ub78c\ub4e4\uc5d0\uac8c 10\ubd84\uac04 \uc0c8\ub85c\uc6b4 \ub9c8\uc778\ud06c\ub798\ud504\ud2b8 \uc6d4\ub4dc\uc5d0\uc11c \uc9d1\uc744 \uc9d3\ub3c4\ub85d \ud55c \ud6c4 \ud574\ub2f9 \ub370\uc774\ud130\ub85c fine-tuning\uc744 \uc9c4\ud589 \ud558\uc600\uc2b5\ub2c8\ub2e4. \ub180\ub78d\uac8c\ub3c4 fine-tuning\ub41c \ubaa8\ub378\uc740 \uae30\uc874 foundation model\uc774 \ud588\ub358 \ud589\ub3d9\uc744 \ub354\uc6b1 \uc798\ud560 \ubfd0\ub9cc \uc544\ub2c8\ub77c \ub098\ubb34\uc640 \ucca0\ub85c \ub41c \uace1\uad2d\uc774\ub3c4 \ub9cc\ub4e4 \uc218 \uc788\uac8c \ub418\uc5c8\uc2b5\ub2c8\ub2e4. <\/p>\n\n\n

\n
\"\"
fine-tuning\uc73c\ub85c \ud5a5\uc0c1\ub418 \uacb0\uacfc\ub97c \ubcf4\uc5ec\uc8fc\ub294 \uadf8\ub798\ud504 (\ucd9c\ucc98: https:\/\/openai.com\/blog\/vpt\/<\/a>, OpenAI \uacf5\uc2dd \ube14\ub85c\uadf8)<\/figcaption><\/figure><\/div>\n\n\n
<\/div>\n\n\n\n

Fine-Tuning with Reinforcement Learning<\/h3>\n\n\n\n

reward\ub97c \uc81c\uacf5\ud558\ub294 \uac15\ud654\ud559\uc2b5 \ubaa8\ub378\uc744 \uc0ac\uc6a9\ud558\uba74 \uc544\ubb34\ub798\ub3c4 \uc870\uae08\ub354 \uace0\ub3c4\ud654\ub41c \uc791\uc5c5\uc744 \ud560 \uc218 \uc788\ub294\ub370\uc694, \uadf8\ub798\uc11c \uc800\uc790\ub294 \ub9c8\uc9c0\ub9c9\uc73c\ub85c fine-tuning\ub41c \ubaa8\ub378\uc5d0 \uac15\ud654\ud559\uc2b5 \uae4c\uc9c0 \uc801\uc6a9\uc744 \uc2dc\ucf1c \ucd5c\uc885\uc801\uc778 \uc131\ub2a5\uc744 \ud14c\uc2a4\ud2b8 \ud574 \ubcf4\uc558\uc2b5\ub2c8\ub2e4. \uc0dd\uac01\ud574 \ubcf4\uba74 \uc774\ub294 \ub525\ub9c8\uc778\ub4dc\uc758 \uc54c\ud30c\uc2a4\ud0c0 \ud559\uc2b5 \ubc29\uc2dd\uacfc \uc720\uc0ac\ud558\ub2e4\uace0 \ubcfc \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc720\uc800\ub4e4\uc758 \ub85c\uadf8\ub85c \uae30\ubcf8\uc801\uc778 \ube4c\ub4dc\uc624\ub354 \ubc0f \ud589\ub3d9\uc591\uc0c1\uc744 \ud559\uc2b5\ud55c \ub2e4\uc74c \uac15\ud654\ud559\uc2b5\uc744 \uc801\uc6a9\ud558\uc5ec \uc778\uac04\ubcf4\ub2e4 \uc798 \ud50c\ub808\uc774 \ud558\ub294 \uc54c\ud30c\uc2a4\ud0c0\ub97c \ud559\uc2b5 \uc2dc\ucf30\uc5c8\uc2b5\ub2c8\ub2e4. \ub9c8\uc778\ud06c\ub798\ud504\ud2b8\uc5d0\uc11c\ub294 \uc81c\uc791 \ub09c\uc774\ub3c4\uac00 \ub9e4\uc6b0 \ub192\uc740 \ub2e4\uc774\uc544\ubaac\ub4dc \uace1\uad2d\uc774\ub97c \uac15\ud654\ud559\uc2b5\uc73c\ub85c \ud559\uc2b5 \uc2dc\ucf30\ub294\ub370\uc694, \uc544\ub798 \uadf8\ub9bc\uc5d0\uc11c \ubcf4\ub4ef\uc774 \ub300\ub7b5 24000\uac1c \uc774\uc0c1\uc758 \ud589\ub3d9\uc774 \ud544\uc694\ud558\uace0 \uc2dc\uac04\ub3c4 20\ubd84 \uc774\uc0c1 \uac78\ub9b0\ub2e4\uace0 \ud569\ub2c8\ub2e4. \ud559\uc2b5\uc744 \uc704\ud574 \ub2e4\uc774\uc544\ubaac\ub4dc \uace1\uad2d\uc774\ub97c \ub9cc\ub4e4\uae30 \uc704\ud55c \uc7ac\ub8cc\ub4e4\uc774 \uc644\uc131\ub420\ub54c \ub9c8\ub2e4 reward\ub97c \uc81c\uacf5 \ud558\uc600\uc2b5\ub2c8\ub2e4. <\/p>\n\n\n\n

\"\"
diamond pickaxe sequence (\ucd9c\ucc98: https:\/\/openai.com\/blog\/vpt\/<\/a>, OpenAI \uacf5\uc2dd \ube14\ub85c\uadf8)<\/figcaption><\/figure>\n\n\n\n

\uac15\ud654\ud559\uc2b5\uc73c\ub85c fine-tuning \ub41c VPT model\uc740 \ub2e4\uc774\uc544\ubaac\ub4dc \uace1\uad2d\uc774\ub97c \ub9cc\ub4e4\uae30 \uc2dc\uc791\ud588\uc744 \ubfd0\ub9cc \uc544\ub2c8\ub77c \uace1\uad2d\uc774\ub97c \ub9cc\ub4e4\uae30 \uc704\ud55c \uc7ac\ub8cc\ub4e4\uc744 \ub9cc\ub4e4 \ub54c\ub3c4 \uc778\uac04\uacfc \ube44\uc2b7\ud55c \uc131\uacf5\ub960\uc744 \ubcf4\uc600\ub2e4\uace0 \ud569\ub2c8\ub2e4. \uc774\ub294 \ub9c8\uc778\ud06c\ub798\ud504\ud2b8\uc5d0\uc11c \ucc98\uc74c\uc73c\ub85c \ub2e4\uc774\uc544\ubaac\ub4dc \uace1\uad2d\uc774 \ub9cc\ub4e4\uae30\uc5d0 \uc131\uacf5\ud55c agent\ub77c\uace0 \ud558\ub124\uc694.<\/p>\n\n\n

\n
\"\"
\uc77c\ubc18\uc801\uc778 \uac15\ud654\ud559\uc2b5\uc73c\ub85c\ub294 \uace1\uad2d\uc774 \uc870\ucc28 \ub9cc\ub4e4\uae30 \ud798\ub4e4\uc5b4 \ubcf4\uc785\ub2c8\ub2e4. (\ucd9c\ucc98: https:\/\/openai.com\/blog\/vpt\/<\/a>, OpenAI \uacf5\uc2dd \ube14\ub85c\uadf8)<\/figcaption><\/figure><\/div>\n\n\n
<\/div>\n\n\n\n

\u7ed3\u8bba<\/h3>\n\n\n\n

\uc774\uc0c1\uc73c\ub85c OpenAI\uc5d0\uc11c \uac1c\ubc1c\ud55c Video PreTraining (VPT)\uc73c\ub85c \ub9c8\uc778\ud06c\ub798\ud504\ud2b8 \ud559\uc2b5\ud558\uae30\uc5d0 \ub300\ud574 \uc54c\uc544 \ubcf4\uc558\uc2b5\ub2c8\ub2e4. \uac8c\uc784 \uc601\uc0c1\uc744 \uac15\ud654\ud559\uc2b5\uc5d0 \uc0ac\uc6a9\ud558\ub294 \ubc29\ubc95\uc744 \uc81c\uc548 \ud558\uc600\uace0, \uc131\ub2a5 \ub610\ud55c \uae30\uc874 \uac15\ud654\ud559\uc2b5\uc5d0\uc11c\ub294 \ubcf4\uc5ec\uc904 \uc218 \uc5c6\ub294 \ud37c\ud3ec\uba3c\uc2a4\ub97c \ubcf4\uc5ec\uc92c\uae30\uc5d0 \ud655\uc2e4\ud788 \ub208\uc5d0 \ub744\ub294 \uc131\uacfc\ub77c \ud560 \uc218 \uc788\uaca0\uc2b5\ub2c8\ub2e4. \ub2e4\ub9cc \uc81c\uac00 \ubcf4\uae30\uc5d0 \uba87\uac00\uc9c0 \ud55c\uacc4\uc810\ub3c4 \ubcf4\uc600\ub294\ub370\uc694, \ub300\ubd80\ubd84\uc758 \ud589\ub3d9 \uc131\uacf5\uc728 \uadf8\ub798\ud504\uac00 log \uadf8\ub798\ud504\ub85c \ub418\uc5b4 \uc788\ub294\ub370, \uc2e4\uc81c \uc131\uacf5\ub960\uc774\ub098 \uc81c\uc791 \ud69f\uc218\ub97c \ud655\uc778\ud574 \ubcf4\uba74 \uadf8\ub807\uac8c \uac12\uc774 \ub192\uc544 \ubcf4\uc774\uc9c4 \uc54a\uc544\uc11c \ud6a8\uacfc\uc801\uc774\ub77c\uace0 \ubcf4\uae30\uc5d0\ub294 \ud798\ub4e4\uaebc \uac19\uc2b5\ub2c8\ub2e4. \ub610\ud55c input\uc73c\ub85c \uc774\ubbf8\uc9c0\ub97c \uc0ac\uc6a9\ud588\ub294\ub370\uc694, \ub300\ubd80\ubd84\uc758 \uac8c\uc784\uc6a9 \uac15\ud654\ud559\uc2b5\uc5d0\uc11c\ub294 image\ub294 \ub178\uc774\uc988\ub3c4 \ub9ce\uace0 \uc815\ubcf4\uac00 \ub108\ubb34 \ubc29\ub300\ud558\uae30\uc5d0 \uc0ac\uc6a9\ud558\uc9c0 \uc54a\uace0, \uac8c\uc784 \uc815\ubcf4\ub97c \ud45c\ud604\ud574 \uc8fc\ub294 scalar \uac12\uc774\ub098 \ubbf8\ub2c8\ub9f5 \ud615\ud0dc\uc758 vector \uac12\uc744 \uc0ac\uc6a9 \ud569\ub2c8\ub2e4. \uc774\ub807\uac8c \uc774\ubbf8\uc9c0\ub97c \uc0ac\uc6a9\ud558\uc9c0 \uc54a\uace0 \ud559\uc2b5\ub41c \uacbd\uc6b0 \uc544\ubb34\ub9ac \uac8c\uc784 \ub3d9\uc601\uc0c1\uc774 \ub9ce\ub354\ub77c\ub3c4 \uc774\ub97c feature\ub85c \ubcc0\ud658\ud558\uae30\ub294 \ud798\ub4dc\ubbc0\ub85c \ud65c\uc6a9\uc131\uc774 \ub2e4\uc18c \ub5a8\uc5b4 \uc9c4\ub2e4\uace0 \ubcfc \uc218\ub3c4 \uc788\uaca0\uc2b5\ub2c8\ub2e4. \uadf8\ub7fc\uc5d0\ub3c4 \ubd88\uad6c\ud558\uace0 \uac8c\uc784 \uc601\uc0c1\uc744 \ud1b5\ud574 \ubcf5\uc7a1\ud55c \ub9c8\uc778\ud06c\ub798\ud504\ud2b8\ub97c \ud559\uc2b5\uc2dc\ucf30\ub2e4\ub294 \uc810\uc5d0\uc11c\ub294 \ub9e4\uc6b0 \ud070 \uc758\ubbf8\uac00 \uc788\ub294 \ub17c\ubb38\uc774\ub77c \uc0dd\uac01\ub429\ub2c8\ub2e4. \uc774 \ub17c\ubb38\uc744 \ubc14\ud0d5\uc73c\ub85c \uac15\ud654\ud559\uc2b5\uc5d0 \ub2e4\uc591\ud55c \uc601\uc0c1 \ud65c\uc6a9 \ubc29\ubc95\uc774 \ub098\uc624\uae38 \uae30\ub300 \ud569\ub2c8\ub2e4. <\/p>\n

<\/span><\/div>","protected":false},"excerpt":{"rendered":"

[\uac00\uc0c1\uc0dd\uba85\uc5f0\uad6c\ud300 \ubc15\uc8fc\ud615] \ucd5c\uadfc \uac15\ud654\ud559\uc2b5\uc5d0\uc11c \ub9ce\uc774 \uc8fc\ubaa9 \ubc1b\uace0 \uc788\ub294 \ubd84\uc57c \uc911\uc5d0 \ud558\ub098\uac00 Offline \ud559\uc2b5 \uc785\ub2c8\ub2e4. \uae30\uc874 \uac15\ud654\ud559\uc2b5\uc740 \uc2dc\ubbac\ub808\uc774\ud130\ub97c \ud1b5\ud574 \uc2e4\uc2dc\uac04\uc73c\ub85c trajectory\ub97c \ubc1b\uc544\uc11c \ud559\uc2b5\uc744 \ud588\uc5c8\uc5b4\uc57c \ud588\ub294\ub370\uc694, \uc2dc\ubbac\ub808\uc774\ud130 \ub9cc\ub4e4\uae30\ub3c4 \uc27d\uc9c0 \uc54a\uace0 \uc2e4\uc81c \uc801\uc6a9\ub3c4 \uc5b4\ub824\uc6cc\uc11c \ub85c\ubd07\uc774\ub098 \uc790\uc728 \uc8fc\ud589 \ucabd\uc5d0\uc11c\ub294 \ub9ce\uc774 \uc2dc\ub3c4 \ub418\ub294 \ubc29\ubc95 \uc785\ub2c8\ub2e4. \uc774\ub97c \uc704\ud574 \uac8c\uc784 \ucabd\uc5d0\uc11c\ub294 \ub2e4\uc591\ud55c \uc720\uc800\ub4e4\uc774 \ub9cc\ub4e4\uc5b4 \ub0b4\ub294 \ub85c\uadf8\ub97c \ud1b5\ud574\uc11c offline reinforcement learning\uc774\ub098 imitation learning\uc744 \uc2dc\ub3c4\ud558\uae30\ub3c4 \ud558\ub294…<\/p>\n

<\/span><\/div>","protected":false},"author":1,"featured_media":63067,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"rank_math_lock_modified_date":false,"footnotes":""},"categories":[566,625,19],"tags":[188,214,427,647,648,649],"class_list":["post-63049","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-gameai","category-rl","category-tech04","tag-featured","tag-reinforcement-learning","tag-openai","tag-vpt","tag-rl","tag-minecraft","category-566","category-625","category-19","description-off"],"_links":{"self":[{"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/posts\/63049","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/comments?post=63049"}],"version-history":[{"count":23,"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/posts\/63049\/revisions"}],"predecessor-version":[{"id":63081,"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/posts\/63049\/revisions\/63081"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/media\/63067"}],"wp:attachment":[{"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/media?parent=63049"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/categories?post=63049"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/smilegate.ai\/cn\/wp-json\/wp\/v2\/tags?post=63049"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}