\n \ub525\ub7ec\ub2dd\uc740 \uc778\uacf5\uc9c0\ub2a5 \ubd84\uc57c\uc5d0\uc11c \uae09\uaca9\ud788 \ubc1c\uc804\ud558\uace0 \uc788\ub294 \uae30\uc220\uc785\ub2c8\ub2e4. \uadf8 \uc911\uc5d0\uc11c\ub3c4 \ud30c\uc774\ud1a0\uce58(PyTorch)\ub294
\n \ub3d9\uc801 \uacc4\uc0b0 \uadf8\ub798\ud504\ub97c \ud2b9\uc9d5\uc73c\ub85c \ud558\uc5ec \uc5f0\uad6c\uc790\uc640 \uac1c\ubc1c\uc790\ub4e4 \uc0ac\uc774\uc5d0\uc11c \uc778\uae30\ub97c \ub04c\uace0 \uc788\uc2b5\ub2c8\ub2e4.
\n \ubcf8 \uae00\uc5d0\uc11c\ub294 \ub525\ub7ec\ub2dd \uc54c\uace0\ub9ac\uc998\uc758 \uae30\ubcf8 \uac1c\ub150, \ud30c\uc774\ud1a0\uce58\uc758 \ud2b9\uc9d5, \uc2e4\uc81c \uc608\uc81c \ucf54\ub4dc\ub97c
\n \ud1b5\ud574 \ub525\ub7ec\ub2dd \ubaa8\ub378\uc744 \uad6c\ud604\ud558\ub294 \ubc29\ubc95\uc744 \uc0c1\uc138\ud558\uac8c \uc124\uba85\ud558\uaca0\uc2b5\ub2c8\ub2e4.\n <\/p>\n
\n \ub525\ub7ec\ub2dd\uc740 \uc778\uacf5 \uc2e0\uacbd\ub9dd\uc744 \uae30\ubc18\uc73c\ub85c \ud55c \uae30\uacc4\ud559\uc2b5\uc758 \ud55c \ubd84\uc57c\ub85c, \uc5ec\ub7ec \uce35\uc73c\ub85c \uad6c\uc131\ub41c \uc2e0\uacbd\ub9dd\uc744
\n \ud1b5\ud574 \ub370\uc774\ud130\ub97c \ucc98\ub9ac\ud558\uace0 \ud559\uc2b5\ud569\ub2c8\ub2e4. \uc2e0\uacbd\ub9dd\uc758 \uac01\uac01\uc758 \uce35\uc740 \uc785\ub825 \ub370\uc774\ud130\uc758 \ud2b9\uc9d5\uc744 \ucd94\ucd9c\ud558\uace0,
\n \uc774\ub97c \ubc14\ud0d5\uc73c\ub85c \ucd5c\uc885\uc801\uc778 \uc608\uce21\uc744 \uc218\ud589\ud569\ub2c8\ub2e4.\n <\/p>\n
\n \uc2e0\uacbd\ub9dd\uc740 \uc785\ub825\uce35, \uc740\ub2c9\uce35, \ucd9c\ub825\uce35\uc73c\ub85c \uad6c\uc131\ub429\ub2c8\ub2e4. \n \uc2e0\uacbd\ub9dd\uc758 \uac01 \ub274\ub7f0\uc5d0\uc11c \uc785\ub825 \uc2e0\ud638\ub97c \ucc98\ub9ac\ud558\uae30 \uc704\ud574 \ud65c\uc131\ud654 \ud568\uc218\uac00 \uc0ac\uc6a9\ub429\ub2c8\ub2e4. \n \ud30c\uc774\ud1a0\uce58\ub294 \ud398\uc774\uc2a4\ubd81\uc774 \uac1c\ubc1c\ud55c \uc624\ud508\uc18c\uc2a4 \ub525\ub7ec\ub2dd \ud504\ub808\uc784\uc6cc\ud06c\ub85c, \ube60\ub978 \ud504\ub85c\ud1a0\ud0c0\uc774\ud551\uacfc \ub3d9\uc801 \uacc4\uc0b0 \uadf8\ub798\ud504\ub97c \n MNIST\ub294 \uc190\uae00\uc528 \uc22b\uc790 \ub370\uc774\ud130\uc14b\uc73c\ub85c, 0\ubd80\ud130 9\uae4c\uc9c0\uc758 \uc22b\uc790\ub97c \ud3ec\ud568\ud55c 70,000\uac1c\uc758 \uc774\ubbf8\uc9c0\ub85c \uad6c\uc131\ub418\uc5b4 \uc788\uc2b5\ub2c8\ub2e4. \uc774\uc81c \uc2e4\uc9c8\uc801\uc73c\ub85c \ud30c\uc774\ud1a0\uce58\ub97c \uc0ac\uc6a9\ud558\uc5ec MNIST \uc22b\uc790 \ubd84\ub958\uae30\ub97c \uad6c\ud604\ud574\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n \n MNIST \ub370\uc774\ud130\uc14b\uc744 \ub2e4\uc6b4\ub85c\ub4dc\ud558\uace0, \ub370\uc774\ud130 \ubcc0\ud658\uc744 \ud1b5\ud574 \ud150\uc11c\ub85c \ubcc0\ud658\ud569\ub2c8\ub2e4.\n <\/p>\n \n \uae30\ubcf8\uc801\uc778 \ub2e4\uce35 \ud37c\uc149\ud2b8\ub860\uc744 \uc815\uc758\ud569\ub2c8\ub2e4.\n <\/p>\n \n \ud559\uc2b5\uc744 \uc704\ud55c \uc190\uc2e4 \ud568\uc218\uc640 \ucd5c\uc801\ud654 \uc0ac\ud56d\uc744 \uc124\uc815\ud558\uace0, \ubaa8\ub378\uc744 \ud559\uc2b5\ud569\ub2c8\ub2e4.\n <\/p>\n \n \ud559\uc2b5\ud55c \ubaa8\ub378\uc744 \ud14c\uc2a4\ud2b8 \ub370\uc774\ud130\uc14b\uc73c\ub85c \ud3c9\uac00\ud569\ub2c8\ub2e4.\n <\/p>\n \n \ubcf8 \uae00\uc5d0\uc11c\ub294 \ud30c\uc774\ud1a0\uce58\ub97c \uc0ac\uc6a9\ud558\uc5ec MNIST \uc190\uae00\uc528 \uc22b\uc790 \ubd84\ub958\uae30\ub97c \uad6c\ud604\ud558\ub294 \uacfc\uc815\uc744 \ud1b5\ud574
\n – \uc785\ub825\uce35<\/strong>: \ub370\uc774\ud130\ub97c \uc785\ub825\ubc1b\ub294 \uce35\uc785\ub2c8\ub2e4.
\n – \uc740\ub2c9\uce35<\/strong>: \uc785\ub825 \ub370\uc774\ud130\ub97c \ubcc0\ud615\ud558\uace0 \ud2b9\uc9d5\uc744 \ucd94\ucd9c\ud558\ub294 \uce35\uc785\ub2c8\ub2e4. \uc5ec\ub7ec \uac1c\uc758 \uc740\ub2c9\uce35\uc744 \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.
\n – \ucd9c\ub825\uce35<\/strong>: \uc608\uce21 \uacb0\uacfc\ub97c \ucd9c\ub825\ud558\ub294 \uce35\uc785\ub2c8\ub2e4.\n <\/p>\n\ud65c\uc131\ud654 \ud568\uc218<\/h3>\n
\n \ub300\ud45c\uc801\uc778 \ud65c\uc131\ud654 \ud568\uc218\ub85c\ub294 ReLU(Rectified Linear Unit), Sigmoid, Tanh \ub4f1\uc774 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n\ud30c\uc774\ud1a0\uce58(PyTorch)<\/h2>\n
\n \uc9c0\uc6d0\ud558\uc5ec \uc720\uc5f0\ud55c \ubaa8\ub378 \uc124\uacc4\ub97c \uac00\ub2a5\ud569\ub2c8\ub2e4. \uc0ac\uc6a9\uc790\uac00 \uc9c1\uad00\uc801\uc73c\ub85c \ubaa8\ub378\uc744 \uad6c\ucd95\ud558\uace0,
\n \uc2e4\ud5d8\ud560 \uc218 \uc788\ub3c4\ub85d \ub3d5\ub294 \ub2e4\uc591\ud55c \uae30\ub2a5\uc744 \uc81c\uacf5\ud569\ub2c8\ub2e4.\n <\/p>\n\ud30c\uc774\ud1a0\uce58\uc758 \uc8fc\uc694 \ud2b9\uc9d5<\/h3>\n
\n
\ub525\ub7ec\ub2dd \uc54c\uace0\ub9ac\uc998 \uad6c\ud604 \uc608\uc81c<\/h2>\n
MNIST \ub370\uc774\ud130\uc14b \uc18c\uac1c<\/h3>\n
\n \uc774 \ub370\uc774\ud130\uc14b\uc740 \ub525\ub7ec\ub2dd \ubaa8\ub378\uc744 \ud3c9\uac00\ud558\ub294 \ub370 \ub110\ub9ac \uc0ac\uc6a9\ub429\ub2c8\ub2e4.\n <\/p>\n\ud30c\uc774\ud1a0\uce58\ub85c MNIST \ubd84\ub958\uae30 \uad6c\ud604\ud558\uae30<\/h3>\n
1. \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac \uc784\ud3ec\ud2b8<\/h4>\n
python\nimport torch\nimport torch.nn as nn\nimport torch.optim as optim\nimport torchvision.transforms as transforms\nimport torchvision.datasets as datasets\nfrom torch.utils.data import DataLoader\n <\/code><\/pre>\n2. \ub370\uc774\ud130\uc14b \ub85c\ub529<\/h4>\n
python\n# \ud558\uc774\ud37c\ud30c\ub77c\ubbf8\ud130 \uc124\uc815\nbatch_size = 64\n\n# \ub370\uc774\ud130 \ubcc0\ud658\ntransform = transforms.Compose([\n transforms.ToTensor(),\n transforms.Normalize((0.5,), (0.5,)) # \ud3c9\uade0\uacfc \ud45c\uc900\ud3b8\ucc28\ub85c \uc815\uaddc\ud654\n])\n\n# \ub370\uc774\ud130\uc14b \ub85c\ub529\ntrain_dataset = datasets.MNIST(root='.\/data', train=True, transform=transform, download=True)\ntrain_loader = DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=True)\n\ntest_dataset = datasets.MNIST(root='.\/data', train=False, transform=transform, download=True)\ntest_loader = DataLoader(dataset=test_dataset, batch_size=batch_size, shuffle=False)\n <\/code><\/pre>\n3. \uc2e0\uacbd\ub9dd \uc815\uc758\ud558\uae30<\/h4>\n
python\nclass SimpleNN(nn.Module):\n def __init__(self):\n super(SimpleNN, self).__init__()\n self.fc1 = nn.Linear(28 * 28, 128) # \uc785\ub825\uce35\n self.fc2 = nn.Linear(128, 64) # \uc740\ub2c9\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 \ud150\uc11c\ub97c 1D \ud150\uc11c\ub85c \ubcc0\ud658\n x = torch.relu(self.fc1(x)) # ReLU \ud65c\uc131\ud654 \ud568\uc218 \uc801\uc6a9\n x = torch.relu(self.fc2(x))\n x = self.fc3(x)\n return x\n <\/code><\/pre>\n4. \ubaa8\ub378 \ud559\uc2b5\ud558\uae30<\/h4>\n
python\n# \ubaa8\ub378, \uc190\uc2e4 \ud568\uc218, \ucd5c\uc801\ud654 \uc54c\uace0\ub9ac\uc998 \uc815\uc758\nmodel = SimpleNN()\ncriterion = nn.CrossEntropyLoss()\noptimizer = optim.Adam(model.parameters(), lr=0.001)\n\n# \ud559\uc2b5 \ub8e8\ud504\nnum_epochs = 5\nfor epoch in range(num_epochs):\n for images, labels in train_loader:\n optimizer.zero_grad() # \uae30\uc6b8\uae30 \ucd08\uae30\ud654\n outputs = model(images) # \uc21c\uc804\ud30c\n loss = criterion(outputs, labels) \n loss.backward() # \uc5ed\uc804\ud30c\n optimizer.step() # \uac00\uc911\uce58 \uc5c5\ub370\uc774\ud2b8\n \n print(f'Epoch [{epoch+1}\/{num_epochs}], Loss: {loss.item():.4f}')\n <\/code><\/pre>\n5. \ubaa8\ub378 \ud3c9\uac00\ud558\uae30<\/h4>\n
python\n# \ubaa8\ub378 \ud3c9\uac00\nmodel.eval() # \ud3c9\uac00 \ubaa8\ub4dc\ub85c \uc804\ud658\nwith torch.no_grad():\n correct = 0\n total = 0\n for images, labels in test_loader:\n outputs = model(images)\n _, predicted = torch.max(outputs.data, 1)\n total += labels.size(0)\n correct += (predicted == labels).sum().item()\n \n print(f'Accuracy of the model on the test images: {100 * correct \/ total:.2f}%')\n <\/code><\/pre>\n\uacb0\ub860<\/h2>\n
\n \ub525\ub7ec\ub2dd\uc758 \uae30\ubcf8 \uac1c\ub150\uacfc \uc54c\uace0\ub9ac\uc998 \uc774\ud574\ub97c \ub3c4\uc654\uc2b5\ub2c8\ub2e4.
\n \ud30c\uc774\ud1a0\uce58\ub294 \uc0ac\uc6a9\uc774 \uac04\ud3b8\ud558\uba74\uc11c\ub3c4 \uac15\ub825\ud55c \uae30\ub2a5\uc744 \uc81c\uacf5\ud558\ubbc0\ub85c, \ub2e4\uc591\ud55c \ub525\ub7ec\ub2dd \ud504\ub85c\uc81d\ud2b8\uc5d0 \uc801\ud569\ud55c
\n \ub3c4\uad6c\uc785\ub2c8\ub2e4. \uc9c0\uc18d\uc801\uc73c\ub85c \ud30c\uc774\ud1a0\uce58\uc758 \ub2e4\uc591\ud55c \uae30\ub2a5\uacfc \ucd5c\uc2e0 \ubaa8\ub378\uc744 \uc2e4\ud5d8\ud558\uc5ec \ub525\ub7ec\ub2dd \uae30\uc220\uc744
\n \ub354\uc6b1 \ubc1c\uc804\uc2dc\ud0a4\uc2dc\uae30 \ubc14\ub78d\ub2c8\ub2e4.\n <\/p>\n\ucc38\uace0 \ubb38\ud5cc<\/h2>\n