{"id":11367,"date":"2025-11-02T17:28:01","date_gmt":"2025-11-02T08:28:01","guid":{"rendered":"https:\/\/www.stechstar.com\/user\/wordpress\/?p=11367"},"modified":"2026-08-20T03:21:07","modified_gmt":"2026-08-19T18:21:07","slug":"%ec%9d%b8%ea%b3%b5%ec%a7%80%eb%8a%a5-%ea%b8%b0%ec%88%a0-%ed%85%90%ec%84%9c%ed%94%8c%eb%a1%9c%ec%9a%b0tensorflow%ec%97%90%ec%84%9c-%ed%8c%8c%ec%9d%b4%ed%86%a0%ec%b9%98pytorch%eb%a1%9c-%ed%99%98","status":"publish","type":"post","link":"https:\/\/www.stechstar.com\/user\/wordpress\/%ec%9d%b8%ea%b3%b5%ec%a7%80%eb%8a%a5-%ea%b8%b0%ec%88%a0-%ed%85%90%ec%84%9c%ed%94%8c%eb%a1%9c%ec%9a%b0tensorflow%ec%97%90%ec%84%9c-%ed%8c%8c%ec%9d%b4%ed%86%a0%ec%b9%98pytorch%eb%a1%9c-%ed%99%98\/","title":{"rendered":"[\uc778\uacf5\uc9c0\ub2a5 \uae30\uc220] \ud150\uc11c\ud50c\ub85c\uc6b0(TensorFlow)\uc5d0\uc11c \ud30c\uc774\ud1a0\uce58(PyTorch)\ub85c \ud658\uc2b9\ud558\uae30"},"content":{"rendered":"

[\uc778\uacf5\uc9c0\ub2a5 \uae30\uc220] \ud150\uc11c\ud50c\ub85c\uc6b0(TensorFlow)\uc5d0\uc11c \ud30c\uc774\ud1a0\uce58(PyTorch)\ub85c \ud658\uc2b9\ud558\uae30<\/h1>\n

*\ud574\ub2f9 \ud3ec\uc2a4\ud305\uc740 \uc774 \uc6d0\ubb38<\/a>\uc744 \ubc88\uc5ed, \uc815\ub9ac(\uc694\uc57d)\ud55c \uac83\uc785\ub2c8\ub2e4.<\/span><\/p>\n

\ud150\uc11c\ud50c\ub85c\uc6b0\uac00 \ud30c\uc774\ud1a0\uce58\ubcf4\ub2e4 1\ub144 \uba3c\uc800 \ub098\uc654\uc9c0\ub9cc, \ucd5c\uadfc \ub9ce\uc740 \uac1c\ubc1c\uc790\ub4e4\uc774 \ud150\uc11c\ud50c\ub85c\uc6b0\uc5d0\uc11c \ud30c\uc774\ud1a0\uce58\ub85c \ub118\uc5b4\uac00\uace0 \uc788\ub294 \uc2e4\uc815\uc774\ub2e4.<\/p>\n

\uc774\ubc88 \uc544\ud2f0\ud074\uc5d0\uc11c\ub294 \ud150\uc11c\ud50c\ub85c\uc6b0\uc5d0\uc11c \ud30c\uc774\ud1a0\uce58\ub85c \ub118\uc5b4\uac00\ub294 \ubc29\ubc95<\/u>\uc5d0 \ub300\ud574 \uc774\uc57c\uae30\ud558\ub824\uace0 \ud55c\ub2e4. <\/p>\n

\uba3c\uc800 \uac01 \ub525\ub7ec\ub2dd \ud504\ub808\uc784\uc6cc\ud06c\ub97c \uc0ac\uc6a9\ud558\ub294 \uc774\uc720\uc5d0 \ub300\ud574 \uc0b4\ud3b4\ubcf8 \ud6c4, \uac04\ub2e8\ud55c \uc608\uc81c\ub97c \ud1b5\ud574 \ud30c\uc774\ud1a0\uce58\uc640 \uac00\uae4c\uc6cc\uc9c0\ub294 \uc2dc\uac04\uc744 \uac00\uc838\ubcf4\ub824 \ud55c\ub2e4.<\/p>\n

1. \ud504\ub808\uc784\uc6cc\ud06c \uac1c\uc694<\/h2>\n

1.1. \ud150\uc11c\ud50c\ub85c\uc6b0<\/h3>\n

2015\ub144 \uad6c\uae00\uc5d0\uc11c \ubc1c\ud45c\ud55c \ub525\ub7ec\ub2dd \ud504\ub808\uc784\uc6cc\ud06c. \ud150\uc11c\ud50c\ub85c\uc6b0 2.0\ub85c \uc5c5\ub370\uc774\ud2b8\ub418\uba74\uc11c \ucf00\ub77c\uc2a4(keras)\uc640 \ud1b5\ud569\ub418\uc5c8\ub2e4.<\/p>\n

1.2. \ud30c\uc774\ud1a0\uce58<\/h3>\n

2016\ub144 \ud398\uc774\uc2a4\ubd81\uc5d0\uc11c \ubc1c\ud45c\ud55c \ub525\ub7ec\ub2dd \ud504\ub808\uc784\uc6cc\ud06c. \ud30c\uc774\uc36c \uc5b8\uc5b4\uc640 \uc790\uc5f0\uc2a4\ub7fd\uac8c \uc5b4\uc6b0\ub7ec\uc9c4\ub2e4\ub294 \uc7a5\uc810\uc774 \uc788\ub2e4.<\/p>\n

\ud30c\uc774\ud1a0\uce58\ub97c \ud150\uc11c \ub2e8\uc704\uc758 \ub525\ub7ec\ub2dd \ubaa8\ub378\uc744 GPU \uc704\uc5d0\uc11c \uacc4\uc0b0\ud560 \uc218 \uc788\ub294 \ud50c\ub7ab\ud3fc<\/u>\uc73c\ub85c \uc774\ud574\ud560 \uc218 \uc788\ub2e4.<\/p>\n

\ub3d9\uc801\uc778(dynamic) \uadf8\ub798\ud504\ub97c \uc0dd\uc131\ud558\uace0, \uadf8\ub54c\uadf8\ub54c(on the fly) \ubaa8\ub378\uc758 \ub3d9\uc791 \uc6d0\ub9ac\ub97c \uc0b4\ud3b4\ubcfc \uc218 \uc788\ub2e4.<\/p>\n

2. \uc65c \ud150\uc11c\ud50c\ub85c\uc6b0\uc5d0\uc11c \ud30c\uc774\ud1a0\uce58\ub85c?<\/h2>\n

– \ud30c\uc774\ud1a0\uce58\uc640 \ub2e4\ub974\uac8c \ud150\uc11c\ud50c\ub85c\uc6b0\ub294 \uc815\uc801\uc778(static) \uadf8\ub798\ud504\ub97c \uc0dd\uc131\ud55c\ub2e4.<\/p>\n

– \ud30c\uc774\ud1a0\uce58\ub294 \ud2b9\uc815 \uadf8\ub798\ud504 \uc791\uc5c5\uc5d0 \ub9de\ucdb0 \ubaa8\ub378\uc744 \uc815\uc758\ud558\uace0 \uc870\uc791\ud560 \uc218 \uc788\ub2e4. (RNN\ucc98\ub7fc \uc785\ub825 \uae38\uc774\uac00 \uac00\ubcc0\uc801\uc778 \uc0c1\ud669\uc5d0\uc11c \uc720\uc6a9)<\/p>\n

– \ud30c\uc774\ud1a0\uce58\ub294 \uadfc\ubcf8\uc801\uc73c\ub85c \ud30c\uc774\uc36c\uc744 \uae30\ubc18\uc73c\ub85c \uac1c\ubc1c\ub418\uc5c8\uae30 \ub54c\ubb38\uc5d0 \uc880\ub354 \uc9c1\uad00\uc801\uc774\ub2e4.<\/p>\n

– \ud150\uc11c\ud50c\ub85c\uc6b0\uac00 \uc775\ud788\uae30 \uc880\ub354 \uae4c\ub2e4\ub85c\uc6b4 \ud3b8. <\/p>\n

– \uc2e0\uc18d\ud558\uac8c \ud504\ub85c\ud1a0\ud0c0\uc785\uc744 \uac1c\ubc1c\ud558\uace0 \uc5f0\uad6c\ud558\ub294 \ub370 \ud30c\uc774\ud1a0\uce58\uac00 \uac00\uc7a5 \uc801\ud569\ud558\ub2e4.<\/p>\n

3. \ud150\uc11c\ud50c\ub85c\uc6b0\uc5d0\uc11c \ud30c\uc774\ud1a0\uce58\ub85c \ub118\uc5b4\uac00\uae30<\/h2>\n

GPU \ud658\uacbd\uc5d0\uc11c \ub450 \ud504\ub808\uc784\uc6cc\ud06c\ub97c \uc124\uce58\ud558\ub294 \uacfc\uc815\uc740 \uc0dd\ub7b5\ud558\uaca0\uc2b5\ub2c8\ub2e4.<\/span><\/p>\n

\ud150\uc11c\ud50c\ub85c\uc6b0\uc5d0\uc11c \ud30c\uc774\ud1a0\uce58\ub85c \ub118\uc5b4\uac00\ub294 \uac74 \uadf8\ub807\uac8c \ubcf5\uc7a1\ud55c \ud3b8\uc774 \uc544\ub2c8\ub2e4.<\/p>\n

\ud30c\uc774\ud1a0\uce58\ub294 \ubb38\uc81c\ub97c \ud574\uacb0\ud558\ub294 \ub370 \uc788\uc5b4 \ud30c\uc774\uc368\ub2c9\ud55c \uc811\uadfc\ubc29\uc2dd<\/i>\uc744 \ucde8\ud558\uae30 \ub54c\ubb38\uc774\ub2e4.<\/p>\n

3.1. \ud150\uc11c \ub2e4\ub8e8\uae30<\/h3>\n

\uac01 \ud504\ub808\uc784\uc6cc\ud06c\uc5d0\uc11c \uc784\uc758\uc758 \ud150\uc11c\ub97c \ucd08\uae30\ud654\ud574\ubcf4\uc790. \ubc29\ubc95\uc740 \ud06c\uac8c \ub2e4\ub974\uc9c0 \uc54a\ub2e4.<\/p>\n

\ud150\uc11c\ud50c\ub85c\uc6b0<\/p>\n

import<\/span> tensorflow as<\/span> tf\n\nrank2_tensor = tf.constant([[1<\/span>, 2<\/span>],\n                            [3<\/span>, 4<\/span>],\n                            [5<\/span>, 6<\/span>]], dtype=tf.int32)<\/code>