\ub525\ub7ec\ub2dd\uc740 \uc778\uacf5\uc9c0\ub2a5(AI)\uc758 \ud55c \ubd84\uc57c\ub85c, \ub2e4\uce35 \uc2e0\uacbd\ub9dd\uc744 \uc0ac\uc6a9\ud558\uc5ec \ub370\uc774\ud130\ub85c\ubd80\ud130 \ud328\ud134\uc744 \ud559\uc2b5\ud558\ub294 \uae30\uc220\uc785\ub2c8\ub2e4.
\n \uc624\ub298\uc740 \uc774\ub7ec\ud55c \ub525\ub7ec\ub2dd\uc5d0\uc11c \uac00\uc7a5 \ub9ce\uc774 \uc0ac\uc6a9\ub418\ub294 \ub450 \uac00\uc9c0 \ud559\uc2b5 \ubc29\ubc95 \uc911 \ud558\ub098\uc778 \uc9c0\ub3c4 \ud559\uc2b5(Supervised Learning)\uc5d0 \ub300\ud574 \uae4a\uc774 \uc788\ub294 \uac15\uc88c\ub97c \uc9c4\ud589\ud558\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
\uc9c0\ub3c4 \ud559\uc2b5\uc740 \uc8fc\uc5b4\uc9c4 \ub370\uc774\ud130\uc5d0 \uae30\ubc18\ud558\uc5ec \uc608\uce21 \ubaa8\ub378\uc744 \ud559\uc2b5\ud558\ub294 \ubc29\ubc95\uc785\ub2c8\ub2e4.
\n \uc5ec\uae30\uc11c ‘\uc9c0\ub3c4’\ub780 \ub808\uc774\ube14\uc774 \uc788\ub294 \ud559\uc2b5 \ub370\uc774\ud130\ub97c \uc758\ubbf8\ud569\ub2c8\ub2e4.
\n \uc9c0\ub3c4 \ud559\uc2b5\uc5d0\uc11c\ub294 \uc785\ub825 \ub370\uc774\ud130\uc640 \ucd9c\ub825 \ub808\uc774\ube14 \uac04\uc758 \uad00\uacc4\ub97c \ud559\uc2b5\ud558\uc5ec
\n \uc0c8\ub85c\uc6b4 \ub370\uc774\ud130\uc5d0 \ub300\ud55c \uc608\uce21\uc744 \ud560 \uc218 \uc788\ub294 \ubaa8\ub378\uc744 \ub9cc\ub4e4\uac8c \ub429\ub2c8\ub2e4.<\/p>\n
\uc9c0\ub3c4 \ud559\uc2b5\uc740 \ud06c\uac8c \ub450 \uac00\uc9c0 \uc720\ud615\uc73c\ub85c \ub098\ub20c \uc218 \uc788\uc2b5\ub2c8\ub2e4: \ubd84\ub958(Classification)\uc640 \ud68c\uadc0(Regression).<\/p>\n
PyTorch\ub294 \ud398\uc774\uc2a4\ubd81\uc774 \uac1c\ubc1c\ud55c \uc624\ud508 \uc18c\uc2a4 \uba38\uc2e0\ub7ec\ub2dd \ub77c\uc774\ube0c\ub7ec\ub9ac\ub85c, \ub525\ub7ec\ub2dd \uc5f0\uad6c\uc790\uc640 \uac1c\ubc1c\uc790\uc5d0\uac8c \uc720\uc6a9\ud55c \ub2e4\uc591\ud55c \uae30\ub2a5\uc744 \uc81c\uacf5\ud569\ub2c8\ub2e4.
\n \ud2b9\ud788, \ub3d9\uc801 \uacc4\uc0b0 \uadf8\ub798\ud504(dynamic computation graph)\ub97c \uc9c0\uc6d0\ud558\uc5ec,
\n \ubaa8\ub378\uc744 \uc27d\uac8c \ub514\ubc84\uae45\ud558\uace0 \uc218\uc815\ud560 \uc218 \uc788\ub294 \uc7a5\uc810\uc774 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
PyTorch\ub97c \uc124\uce58\ud558\ub824\uba74 \ub2e4\uc74c\uc758 \uba85\ub839\uc5b4\ub97c \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.
\n \uc544\ub798 \uba85\ub839\uc5b4\ub294 pip\ub97c \uc774\uc6a9\ud558\uc5ec PyTorch\ub97c \uc124\uce58\ud558\ub294 \ubc29\ubc95\uc785\ub2c8\ub2e4:<\/p>\n
pip install torch torchvision torchaudio<\/code><\/pre>\n3. \ub525\ub7ec\ub2dd \ubaa8\ub378 \ub9cc\ub4e4\uae30<\/h2>\n
\uc774\uc81c PyTorch\ub97c \uc0ac\uc6a9\ud558\uc5ec \uac04\ub2e8\ud55c \ub525\ub7ec\ub2dd \ubaa8\ub378\uc744 \ub9cc\ub4e4\uc5b4 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4.
\n \uc774\ubc88 \uc608\uc81c\ub294 \ubd84\ub958 \ubb38\uc81c\ub97c \ub2e4\ub8e8\uba70, \uc720\uba85\ud55c MNIST \ub370\uc774\ud130\uc14b\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc190\uae00\uc528 \uc22b\uc790\ub97c \ubd84\ub958\ud558\ub294 \ubaa8\ub378\uc744 \uad6c\ucd95\ud560 \uac83\uc785\ub2c8\ub2e4.
\n MNIST \ub370\uc774\ud130\uc14b\uc740 0\ubd80\ud130 9\uae4c\uc9c0\uc758 \uc22b\uc790 \uc774\ubbf8\uc9c0\ub85c \uad6c\uc131\ub41c \ub370\uc774\ud130\uc14b\uc785\ub2c8\ub2e4.<\/p>\n
3.1 \ub370\uc774\ud130\uc14b \ub85c\ub529<\/h3>\n
\uba3c\uc800 MNIST \ub370\uc774\ud130\uc14b\uc744 \ub85c\ub529\ud558\uace0, \uc774\ub97c \ud559\uc2b5 \ubc0f \ud14c\uc2a4\ud2b8 \ub370\uc774\ud130\ub85c \ub098\ub204\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
import torch\nfrom torchvision import datasets, transforms\n\n# \ub370\uc774\ud130 \ubcc0\ud658 (\uc815\uaddc\ud654)\ntransform = transforms.Compose([\n transforms.ToTensor(),\n transforms.Normalize((0.5,), (0.5,)) # \ud3c9\uade0 0.5, \ud45c\uc900\ud3b8\ucc28 0.5\ub85c \uc815\uaddc\ud654\n])\n\n# MNIST \ub370\uc774\ud130\uc14b \ub2e4\uc6b4\ub85c\ub4dc \ubc0f \ub85c\ub529\ntrain_dataset = datasets.MNIST(root='.\/data', train=True, download=True, transform=transform)\ntest_dataset = datasets.MNIST(root='.\/data', train=False, download=True, transform=transform)\n\ntrain_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=64, shuffle=True)\ntest_loader = torch.utils.data.DataLoader(dataset=test_dataset, batch_size=64, shuffle=False)<\/code><\/pre>\n3.2 \ubaa8\ub378 \uc815\uc758<\/h3>\n
\ub525\ub7ec\ub2dd \ubaa8\ub378\uc740 nn.Module\uc744 \uc0c1\uc18d\ubc1b\uc544 \uc815\uc758\ud569\ub2c8\ub2e4. \uc5ec\uae30\uc11c \uac04\ub2e8\ud55c \uc2e0\uacbd\ub9dd \ubaa8\ub378\uc744 \uc815\uc758\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
import torch.nn as nn\nimport torch.nn.functional as F\n\nclass SimpleNN(nn.Module):\n def __init__(self):\n super(SimpleNN, self).__init__()\n self.fc1 = nn.Linear(28 * 28, 128) # \uccab \ubc88\uc9f8 \uce35\n self.fc2 = nn.Linear(128, 64) # \ub450 \ubc88\uc9f8 \uce35\n self.fc3 = nn.Linear(64, 10) # \ucd9c\ub825 \uce35\n\n def forward(self, x):\n x = x.view(-1, 28 * 28) # 2D \uc774\ubbf8\uc9c0\ub97c 1D \ubca1\ud130\ub85c \ubcc0\ud658\n x = F.relu(self.fc1(x)) # \ud65c\uc131\ud654 \ud568\uc218 ReLU \uc801\uc6a9\n x = F.relu(self.fc2(x))\n x = self.fc3(x)\n return x\n\n# \ubaa8\ub378 \uc778\uc2a4\ud134\uc2a4 \uc0dd\uc131\nmodel = SimpleNN()<\/code><\/pre>\n3.3 \uc190\uc2e4 \ud568\uc218 \ubc0f \ucd5c\uc801\ud654\uae30 \uc815\uc758<\/h3>\n
\uc774\uc81c \uc190\uc2e4 \ud568\uc218\uc640 \ucd5c\uc801\ud654\uae30\ub97c \uc815\uc758\ud558\uaca0\uc2b5\ub2c8\ub2e4. \uc6b0\ub9ac\ub294 \uad50\ucc28 \uc5d4\ud2b8\ub85c\ud53c \uc190\uc2e4 \ud568\uc218\uc640 Adam \ucd5c\uc801\ud654\uae30\ub97c \uc0ac\uc6a9\ud558\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
import torch.optim as optim\n\n# \uc190\uc2e4 \ud568\uc218\ncriterion = nn.CrossEntropyLoss()\n# \ucd5c\uc801\ud654\uae30\noptimizer = optim.Adam(model.parameters(), lr=0.001)<\/code><\/pre>\n3.4 \ubaa8\ub378 \ud6c8\ub828<\/h3>\n
\uc774\uc81c \ubaa8\ub378\uc744 \ud559\uc2b5\uc2dc\ud0ac \uc2dc\uac04\uc785\ub2c8\ub2e4. \uac01 \uc5d0\ud3ed(epoch)\ub9c8\ub2e4 \ub370\uc774\ud130\ub97c \ubc18\ubcf5\uc801\uc73c\ub85c \ubaa8\ub378\uc5d0 \ud1b5\uacfc\uc2dc\ud0a4\uace0,
\n \uc190\uc2e4\uc744 \uacc4\uc0b0\ud558\uc5ec \uac00\uc911\uce58\ub97c \uc5c5\ub370\uc774\ud2b8\ud569\ub2c8\ub2e4.<\/p>\n
num_epochs = 5\n\nfor epoch in range(num_epochs):\n for images, labels in train_loader:\n # \uae30\uc6b8\uae30 \ucd08\uae30\ud654\n optimizer.zero_grad()\n # \ubaa8\ub378\uc5d0 \uc774\ubbf8\uc9c0 \ud1b5\uacfc\n outputs = model(images)\n # \uc190\uc2e4 \uacc4\uc0b0\n loss = criterion(outputs, labels)\n # \uc5ed\uc804\ud30c\n loss.backward()\n # \uac00\uc911\uce58 \uc5c5\ub370\uc774\ud2b8\n optimizer.step()\n \n print(f'Epoch [{epoch+1}\/{num_epochs}], Loss: {loss.item():.4f}')<\/code><\/pre>\n3.5 \ubaa8\ub378 \ud3c9\uac00<\/h3>\n
\ud559\uc2b5\uc774 \uc644\ub8cc\ub418\uba74, \ud14c\uc2a4\ud2b8 \ub370\uc774\ud130\ub97c \uc0ac\uc6a9\ud558\uc5ec \ubaa8\ub378\uc744 \ud3c9\uac00\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
correct = 0\ntotal = 0\n\nwith torch.no_grad(): # \uae30\uc6b8\uae30 \uacc4\uc0b0 \ube44\ud65c\uc131\ud654\n for images, labels in test_loader:\n outputs = model(images)\n _, predicted = torch.max(outputs.data, 1) # \uac00\uc7a5 \ud070 \uac12\uc758 \uc778\ub371\uc2a4 \ucd94\ucd9c\n total += labels.size(0)\n correct += (predicted == labels).sum().item()\n\nprint(f'Accuracy of the model on the test images: {100 * correct \/ total:.2f}%')<\/code><\/pre>\n4. \uacb0\ub860<\/h2>\n
\uc774\ubc88 \uac15\uc88c\uc5d0\uc11c\ub294 PyTorch\ub97c \uc0ac\uc6a9\ud558\uc5ec \uae30\ubcf8\uc801\uc778 \ub525\ub7ec\ub2dd \ubaa8\ub378\uc744 \uad6c\ucd95\ud558\uace0,
\n MNIST \ub370\uc774\ud130\uc14b\uc744 \ud1b5\ud574 \ubd84\ub958 \ubb38\uc81c\ub97c \ud574\uacb0\ud558\ub294 \ubc29\ubc95\uc744 \ubc30\uc6e0\uc2b5\ub2c8\ub2e4.
\n \uc9c0\ub3c4 \ud559\uc2b5\uc744 \ud65c\uc6a9\ud55c \uc774 \uc811\uadfc\ubc95\uc740 \ub2e4\uc591\ud55c \uc751\uc6a9 \ubd84\uc57c\uc5d0 \uc751\uc6a9\ub420 \uc218 \uc788\uc73c\uba70,
\n \ub354 \ubcf5\uc7a1\ud55c \ubaa8\ub378\ub85c \ubc1c\uc804\uc2dc\ud0ac \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
4.1 \ucd94\uac00 \ud559\uc2b5 \uc790\ub8cc<\/h3>\n
\ub354 \uae4a\uc774 \uc788\ub294 \ub525\ub7ec\ub2dd\uc744 \ubc30\uc6b0\uace0 \uc2f6\ub2e4\uba74 \ub2e4\uc74c\uc758 \uc790\ub8cc\ub97c \ucc38\uace0\ud574 \ubcf4\uc138\uc694:<\/p>\n
\n- Deep Learning by Ian Goodfellow et al.<\/li>\n
- \ud30c\uc774\ud1a0\uce58 \uacf5\uc2dd \ubb38\uc11c: https:\/\/pytorch.org\/docs\/stable\/index.html<\/a><\/li>\n
- CS231n: Convolutional Neural Networks for Visual Recognition<\/li>\n<\/ul>\n
\ub525\ub7ec\ub2dd\uc740 \ubc29\ub300\ud55c \uc5f0\uad6c \ubd84\uc57c\ub85c, \uc9c0\uc18d\uc801\uc778 \ud559\uc2b5\uc774 \ud544\uc694\ud569\ub2c8\ub2e4.
\n \uc5ec\ub7ec\ubd84\uc758 \ub525\ub7ec\ub2dd \uc5ec\uc815\uc5d0 \ub3c4\uc6c0\uc774 \ub418\uae38 \ubc14\ub78d\ub2c8\ub2e4!<\/p>\n
<\/body><\/p>\n","protected":false},"excerpt":{"rendered":"