{"id":40444,"date":"2024-11-04T01:51:47","date_gmt":"2024-11-04T01:51:47","guid":{"rendered":"https:\/\/atmokpo.com\/w\/?p=40444"},"modified":"2024-11-26T06:39:10","modified_gmt":"2024-11-26T06:39:10","slug":"5-%ea%b0%95%ed%99%94%ed%95%99%ec%8a%b5%ec%9d%98-%ec%97%ad%ec%82%ac%ec%99%80-%ec%a3%bc%ec%9a%94-%ec%9d%91%ec%9a%a9-%eb%b6%84%ec%95%bc-rl%ec%9d%98-%eb%b0%9c%ec%a0%84-%ea%b3%bc%ec%a0%95","status":"publish","type":"post","link":"https:\/\/atmokpo.com\/w\/40444\/","title":{"rendered":"5. \uac15\ud654\ud559\uc2b5\uc758 \uc5ed\uc0ac\uc640 \uc8fc\uc694 \uc751\uc6a9 \ubd84\uc57c, RL\uc758 \ubc1c\uc804 \uacfc\uc815"},"content":{"rendered":"

<\/p>\n

1. \uc11c\ub860<\/h2>\n

\uac15\ud654\ud559\uc2b5(Reinforcement Learning, RL)\uc740 Machine Learning\uc758 \ud55c \ubd84\uc57c\ub85c, \uc5d0\uc774\uc804\ud2b8\uac00 \ud658\uacbd\uacfc \uc0c1\ud638\uc791\uc6a9\ud558\uba74\uc11c \uc2a4\uc2a4\ub85c \ucd5c\uc801\uc758 \ud589\ub3d9\uc744 \ud559\uc2b5\ud558\ub294 \uae30\uc220\uc785\ub2c8\ub2e4. \uc774 \uacfc\uc815\uc5d0\uc11c \uc5d0\uc774\uc804\ud2b8\ub294 \ubcf4\uc0c1\uc744 \ucd5c\ub300\ud654\ud558\ub824\uace0 \ub178\ub825\ud558\uba70, \uc774\ub97c \ud1b5\ud574 \uc815\ucc45\uc744 \ucd5c\uc801\ud654\ud558\ub294 \ubc29\uc2dd\uc785\ub2c8\ub2e4. \ubcf8 \uae00\uc5d0\uc11c\ub294 \uac15\ud654\ud559\uc2b5\uc758 \uc5ed\uc0ac, \uc8fc\uc694 \uc751\uc6a9 \ubd84\uc57c, \uadf8\ub9ac\uace0 \ucd5c\uadfc \ubc1c\uc804 \uacfc\uc815\uc744 \ub2e4\ub8f0 \uac83\uc785\ub2c8\ub2e4.<\/p>\n

2. \uac15\ud654\ud559\uc2b5\uc758 \uc5ed\uc0ac<\/h2>\n

2.1 \ucd08\uae30 \uc5f0\uad6c (1950-1980\ub144\ub300)<\/h3>\n

\uac15\ud654\ud559\uc2b5\uc758 \uae30\uc6d0\uc740 1950\ub144\ub300 \ud6c4\ubc18\uc73c\ub85c \uac70\uc2ac\ub7ec \uc62c\ub77c\uac11\ub2c8\ub2e4. \uc2ec\ub9ac\ud559\uc758 \ud589\ub3d9 \uc774\ub860\uc5d0 \uae30\ubc18\uc744 \ub450\uace0 \uc788\uc73c\uba70, \ud2b9\ud788 \ud5e8\ub9ac \uc0c1\uad00 \uc774\ub860<\/strong>\uacfc \ud30c\ube14\ub85c\ud504\uc758 \uac1c \uc774\ub860<\/strong>\uc774 \uc911\uc694\ud55c \uc601\ud5a5\uc744 \ubbf8\ucce4\uc2b5\ub2c8\ub2e4. \ucd08\uae30\uc758 \uac15\ud654\ud559\uc2b5 \uc54c\uace0\ub9ac\uc998\uc740 \uc8fc\ub85c \uc815\ucc45 \uacbd\uc0ac \uae30\ubc95(Policy Gradient methods)<\/code>\uc640 Q-\ub7ec\ub2dd(Q-learning)<\/code> \ub4f1\uc758 \uac04\ub2e8\ud55c \ubc29\ubc95\ub860\uc5d0\uc11c \ucd9c\ubc1c\ud588\uc2b5\ub2c8\ub2e4.<\/p>\n

2.2 1980\ub144\ub300\uc758 \ubc1c\uc804<\/h3>\n

1980\ub144\ub300\uc5d0 \ub4e4\uc5b4\uc11c\ub294 \ubb38\uc81c \ud574\uacb0\uc744 \uc704\ud55c \uc54c\uace0\ub9ac\uc998 \uac1c\ubc1c\uc774 \uc774\ub8e8\uc5b4\uc84c\uc2b5\ub2c8\ub2e4. \ud2b9\ud788 \ub4dc\ub85c\ub974\ud504\uc758 \uac15\ud654\ud559\uc2b5 \uc774\ub860<\/strong>\uacfc \ub9c8\ub974\ucf54\ud504 \uacb0\uc815 \uacfc\uc815(Markov Decision Processes, MDP)<\/strong>\uc758 \uad00\uacc4\uac00 \uba85\ud655\ud574\uc9c0\uba74\uc11c, \uac15\ud654\ud559\uc2b5\uc758 \uc218\ud559\uc801 \uae30\ucd08\uac00 \ub2e6\uc774\uac8c \ub429\ub2c8\ub2e4. \ub2f9\uc2dc\uc5d0\ub294 \ub9ce\uc740 \uc8fc\uc694 \uc5f0\uad6c\ub4e4\uc774 \uc774\ub7ec\ud55c \ubd84\uc57c\uc5d0\uc11c \uc774\ub8e8\uc5b4\uc84c\uc2b5\ub2c8\ub2e4.<\/p>\n

2.3 1990\ub144\ub300\uc758 \ub3c4\uc57d<\/h3>\n

1990\ub144\ub300 \uc911\ubc18, \ub525 \ub7ec\ub2dd<\/strong>\uacfc \ud569\uccd0\uc9c4 \uac15\ud654\ud559\uc2b5\uc774 \ub4f1\uc7a5\ud558\uba74\uc11c \uc774 \ubd84\uc57c\ub294 \uae09\uaca9\ud55c \ubc1c\uc804\uc744 \ud558\uac8c \ub429\ub2c8\ub2e4. \ud2b9\ud788 Q-Learning<\/code> \uc54c\uace0\ub9ac\uc998\uc758 \ubc1c\uc804\uc740 \ub2e4\uc591\ud55c \ubb38\uc81c \ud574\uacb0\uc744 \uc704\ud55c \uac15\ub825\ud55c \ub3c4\uad6c\ub85c \uc790\ub9ac \uc7a1\uc558\uc2b5\ub2c8\ub2e4. \uc774 \uc2dc\uae30\uc5d0 \ub2e4\uc591\ud55c \uc54c\uace0\ub9ac\uc998\uacfc \uc774\ub860<\/strong>\uc774 \uac1c\ubc1c\ub418\uba74\uc11c \ub9ce\uc740 \uc5f0\uad6c\uc790\ub4e4\uc774 \uc774 \ubd84\uc57c\uc5d0 \uad00\uc2ec\uc744 \uac00\uc9c0\uae30 \uc2dc\uc791\ud588\uc2b5\ub2c8\ub2e4.<\/p>\n

2.4 2000\ub144\ub300 \uc774\ud6c4<\/h3>\n

2000\ub144\ub300\ubd80\ud130 \uac15\ud654\ud559\uc2b5\uc740 \uc778\uacf5\uc9c0\ub2a5 \ubd84\uc57c\uc5d0\uc11c \uc911\uc2ec\uc801\uc778 \uc5ed\ud560\uc744 \ud558\uac8c \ub429\ub2c8\ub2e4. \ub525 Q-\ub7ec\ub2dd<\/strong>\uacfc \uac19\uc740 \uc54c\uace0\ub9ac\uc998\uc774 \ubc1c\uc804\ud558\uba74\uc11c \ube44\ub514\uc624 \uac8c\uc784, \ub85c\ubd07 \uc81c\uc5b4 \ub4f1 \ub2e4\uc591\ud55c \ubd84\uc57c\uc5d0\uc11c \ud65c\uc6a9\ub429\ub2c8\ub2e4. \ud2b9\ud788, 2015\ub144 DeepMind<\/strong>\uc758 AlphaGo<\/code>\ub294 \ubc14\ub451\uc5d0\uc11c \uc778\uac04 \ucd5c\uace0\uc218\ub97c \uc774\uacbc\uace0, \uc774\ub294 \uac15\ud654\ud559\uc2b5\uc758 \ub300\uc911\ud654\uc5d0 \uae30\uc5ec\ud588\uc2b5\ub2c8\ub2e4.<\/p>\n

3. \uac15\ud654\ud559\uc2b5\uc758 \uc8fc\uc694 \uc751\uc6a9 \ubd84\uc57c<\/h2>\n

\uac15\ud654\ud559\uc2b5\uc740 \ub2e4\uc591\ud55c \ubd84\uc57c\uc5d0\uc11c \ud601\uc2e0\uacfc \ubb38\uc81c \ud574\uacb0\uc5d0 \uae30\uc5ec\ud558\uace0 \uc788\uc2b5\ub2c8\ub2e4. \ub2e4\uc74c\uc740 \uadf8 \uc8fc\uc694 \uc751\uc6a9 \ubd84\uc57c\uc785\ub2c8\ub2e4.<\/p>\n

3.1 \uac8c\uc784<\/h3>\n

\uac15\ud654\ud559\uc2b5\uc740 \uac8c\uc784 AI \uac1c\ubc1c\uc5d0 \ub110\ub9ac \uc0ac\uc6a9\ub429\ub2c8\ub2e4. AlphaGo<\/strong>\uc640 \uac19\uc740 \ud504\ub85c\uadf8\ub7a8\uc740 \uac15\ud654\ud559\uc2b5\uc744 \uc774\uc6a9\ud558\uc5ec \uc804\ub7b5\uc801\uc778 \uacb0\uc815\uc744 \ub0b4\ub9ac\uba70, \uc2a4\ud0c0\ud06c\ub798\ud504\ud2b8\uc640 \uac19\uc740 \ubcf5\uc7a1\ud55c \uac8c\uc784\uc5d0\uc11c\ub3c4 \uc131\uacf5\uc744 \uac70\ub450\uace0 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n

3.2 \ub85c\ubd07 \uacf5\ud559<\/h3>\n

\ub85c\ubd07 \uc81c\uc5b4 \ubd84\uc57c\uc5d0\uc11c\ub3c4 \uac15\ud654\ud559\uc2b5\uc774 \ud65c\ubc1c\ud788 \uc5f0\uad6c\ub418\uace0 \uc788\uc2b5\ub2c8\ub2e4. \ub85c\ubd07\uc740 \uac15\ud654\ud559\uc2b5\uc744 \ud1b5\ud574 \ud658\uacbd\uc744 \ud0d0\uc0c9\ud558\uace0, \ucd5c\uc801\uc758 \uacbd\ub85c\ub97c \uacc4\ud68d\ud558\uba70, \ube44\ud6a8\uc728\uc801\uc778 \ub3d9\uc791\uc744 \ucd5c\uc18c\ud654\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n

3.3 \uc790\uc728 \uc8fc\ud589<\/h3>\n

\uc790\uc728 \uc8fc\ud589\ucc28\ub294 \uac15\ud654\ud559\uc2b5\uc744 \ud1b5\ud574 \ub3c4\ub85c \uc8fc\ud589 \uc911 \ub2e4\uc591\ud55c \uc0c1\ud669\uc744 \ud559\uc2b5\ud558\uba70, \uc774\ub97c \ud1b5\ud574 \uc548\uc804\ud55c \uc8fc\ud589 \uacbd\ub85c\ub97c \uc120\ud0dd\ud569\ub2c8\ub2e4. \uac15\ud654\ud559\uc2b5\uc740 \uc2e4\uc81c \uc8fc\ud589 \uc0c1\ud669\uc5d0\uc11c\ub3c4 \uc77c\uad00\ub418\uace0 \ube60\ub978 \uacb0\uc815\uc744 \ub0b4\ub9ac\ub294 \ub370 \ub3c4\uc6c0\uc744 \uc90d\ub2c8\ub2e4.<\/p>\n

3.4 \uae08\uc735<\/h3>\n

\uae08\uc735 \ubd84\uc57c\uc5d0\uc11c\ub294 \uac15\ud654\ud559\uc2b5\uc744 \uc774\uc6a9\ud55c \uc790\ub3d9 \uac70\ub798 \uc2dc\uc2a4\ud15c\uc774 \uc5f0\uad6c\ub418\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc5d0\uc774\uc804\ud2b8\ub294 \uc8fc\uc2dd \uc2dc\uc7a5\uc5d0\uc11c \uc774\uc775\uc744 \uadf9\ub300\ud654\ud558\uae30 \uc704\ud574 \ub2e4\uc591\ud55c \uc804\ub7b5\uc744 \uc2e4\ud5d8\ud558\uace0 \ud559\uc2b5\ud569\ub2c8\ub2e4.<\/p>\n

3.5 \uc758\ub8cc<\/h3>\n

\uc758\ub8cc \ubd84\uc57c\uc5d0\uc11c\ub294 \ud658\uc790\uc758 \uce58\ub8cc \ubc29\uc548\uc744 \uacb0\uc815\ud558\ub294 \ub370 \uac15\ud654\ud559\uc2b5\uc774 \uc0ac\uc6a9\ub429\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, \uac1c\uc778 \ub9de\ucda4\ud615 \uce58\ub8cc \ubc0f \uc57d\ubb3c \ubc30\uae09\uc5d0\uc11c \ucd5c\uc801\uc758 \uacb0\uc815\uc744 \ub0b4\ub9ac\ub294 \ub370 \uae30\uc5ec\ud558\uace0 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n

4. RL\uc758 \ubc1c\uc804 \uacfc\uc815<\/h2>\n

\uac15\ud654\ud559\uc2b5\uc758 \ubc1c\uc804 \uacfc\uc815\uc740 \ub2e4\uc74c\uacfc \uac19\uc740 \uc8fc\uc694 \ub2e8\uacc4\ub85c \uad6c\uc131\ub429\ub2c8\ub2e4.<\/p>\n

4.1 \uc774\ub860\uc801 \uae30\ucd08<\/h3>\n

\uac15\ud654\ud559\uc2b5\uc758 \uc774\ub860\uc801 \uae30\ucd08\uac00 \ud655\ub9bd\ub418\uba74\uc11c, \ub2e4\uc591\ud55c \uc54c\uace0\ub9ac\uc998\uc774 \ubc1c\uc0dd\ud558\uae30 \uc2dc\uc791\ud588\uc2b5\ub2c8\ub2e4. MDP<\/code>, \ubca8\ub9cc \ubc29\uc815\uc2dd(Bellman Equation)<\/code> \ub4f1\uc774 \uc774\ub7ec\ud55c \uae30\ucd08\uc758 \uc608\uc785\ub2c8\ub2e4.<\/p>\n

4.2 \uc2e4\uc6a9 \uc54c\uace0\ub9ac\uc998\uc758 \uac1c\ubc1c<\/h3>\n

\uc2e4\uc81c \ubb38\uc81c \ud574\uacb0\uc744 \uc704\ud55c \ub2e4\uc591\ud55c \uc54c\uace0\ub9ac\uc998\uc774 \uac1c\ubc1c\ub418\uc5c8\uc2b5\ub2c8\ub2e4. A3C(Asynchronous Actor-Critic)<\/code>, DDPG(Deep Deterministic Policy Gradient)<\/code> \ub4f1\uc758 \uc54c\uace0\ub9ac\uc998\uc774 \ubc1c\uc804\ud558\uba74\uc11c \uc131\ub2a5\uc774 \ud06c\uac8c \ud5a5\uc0c1\ub418\uc5c8\uc2b5\ub2c8\ub2e4.<\/p>\n

4.3 HRL(Hierarchical Reinforcement Learning)<\/h3>\n

\uc7ac\uc0ac\uc6a9 \uac00\ub2a5\ud55c \ud558\uc704 \uc815\ucc45\uc758 \ud559\uc2b5\uc744 \ud1b5\ud574 \uac15\ud654\ud559\uc2b5\uc758 \ud6a8\uc728\uc744 \uc99d\uac00\uc2dc\ud0a4\ub294 \ubc29\ud5a5\uc73c\ub85c HRL\uc774 \ubc1c\uc804\ud558\uc600\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \ubc29\ubc95\ub860\uc740 \ubcf5\uc7a1\ud55c \ubb38\uc81c\ub97c \ub2e8\uc21c\ud654\ud558\ub294 \ub370 \ud070 \ub3c4\uc6c0\uc774 \ub429\ub2c8\ub2e4.<\/p>\n

4.4 \ucd5c\uadfc\uc758 \uc9c4\uc804<\/h3>\n

\ucd5c\uadfc\uc758 \uac15\ud654\ud559\uc2b5\uc740 \ub525 \ub7ec\ub2dd\uacfc\uc758 \uacb0\ud569\uc744 \ud1b5\ud574 \ube44\uc57d\uc801\uc778 \uc131\uc7a5\uc744 \uc774\ub8e8\uc5c8\uc2b5\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4 AlphaFold<\/code>\uc640 \uac19\uc740 \uc5f0\uad6c\ub4e4\uc740 \uc0dd\uba85 \uacfc\ud559 \ubb38\uc81c\ub97c \ud574\uacb0\ud558\uae30 \uc704\ud574 \uac15\ud654\ud559\uc2b5\uacfc \uc2e0\uacbd\ub9dd\uc744 \uacb0\ud569\ud588\uc2b5\ub2c8\ub2e4.<\/p>\n

5. \uacb0\ub860<\/h2>\n

\uac15\ud654\ud559\uc2b5\uc740 \uc778\uacf5\uc9c0\ub2a5 \ubd84\uc57c\uc5d0\uc11c \ube60\ub974\uac8c \ubc1c\uc804\ud558\uace0 \uc788\uc73c\uba70, \ub2e4\uc591\ud55c \ubd84\uc57c\uc5d0\uc11c \uadf8 \ud6a8\uc6a9\uc131\uc744 \uc785\uc99d\ud558\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc55e\uc73c\ub85c\ub3c4 \uc774 \uae30\uc220\uc740 \ub354\uc6b1 \ubc1c\uc804\ud558\uc5ec \ub2e4\uc591\ud55c \ubd84\uc57c\uc5d0\uc11c \ud601\uc2e0\uc744 \uc77c\uc73c\ud0ac \uac83\uc73c\ub85c \uae30\ub300\ub429\ub2c8\ub2e4. \ubcf8 \uae00\uc744 \ud1b5\ud574 \ub3c5\uc790 \uc5ec\ub7ec\ubd84\uc774 \uac15\ud654\ud559\uc2b5\uc758 \uc5ed\uc0ac\uc640 \uc8fc\uc694 \uc751\uc6a9 \ubd84\uc57c\uc5d0 \ub300\ud574 \uc774\ud574\ud558\ub294 \ub370 \ub3c4\uc6c0\uc774 \ub418\uc5c8\uae30\ub97c \ubc14\ub78d\ub2c8\ub2e4.<\/p>\n

\ucc38\uace0\ubb38\ud5cc<\/h2>\n