\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, \ub370\uc774\ud130\ub85c\ubd80\ud130 \ud328\ud134\uc744 \ud559\uc2b5\ud558\uc5ec \uc608\uce21 \ub610\ub294 \ubd84\ub958 \ub4f1\uc758 \ud0dc\uc2a4\ud06c\ub97c \uc218\ud589\ud558\ub294 \ub370 \uc0ac\uc6a9\ub429\ub2c8\ub2e4. \ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 \ud30c\uc774\ud1a0\uce58(PyTorch)\ub77c\ub294 \ub525\ub7ec\ub2dd \ud504\ub808\uc784\uc6cc\ud06c\ub97c \uc0ac\uc6a9\ud558\uc5ec \ub525\ub7ec\ub2dd\uc758 \uae30\ubcf8 \uac1c\ub150\uacfc \ud568\uaed8 \ud559\uc2b5 \uc54c\uace0\ub9ac\uc998\uc5d0 \ub300\ud574 \uc790\uc138\ud788 \uc124\uba85\ud558\uaca0\uc2b5\ub2c8\ub2e4.\n<\/p>\n
\n \ub525\ub7ec\ub2dd\uc758 \ud575\uc2ec\uc740 \uc2e0\uacbd\ub9dd(Neural Networks)\uc785\ub2c8\ub2e4. \uc2e0\uacbd\ub9dd\uc740 \ub178\ub4dc(Node)\ub77c\uace0 \ubd88\ub9ac\ub294 \ub2e8\uc704\ub4e4\uc774 \uce35\uc744 \uc774\ub8e8\uc5b4 \uc5f0\uacb0\ub41c \uad6c\uc870\ub85c, \uc785\ub825 \ub370\uc774\ud130\ub97c \ubc1b\uace0 \uac00\uc911\uce58(weight)\uc640 \ud3b8\ud5a5(bias)\uc744 \uc801\uc6a9\ud558\uc5ec \ucd9c\ub825 \ub370\uc774\ud130\ub97c \uc0dd\uc131\ud569\ub2c8\ub2e4.
\n \uac01 \ub178\ub4dc\ub294 \ube44\uc120\ud615 \ubcc0\ud658\uc744 \uc218\ud589\ud558\uace0, \uc774\ub7ec\ud55c \ubcc0\ud658\uc740 \ud65c\uc131\ud654 \ud568\uc218(Activation Function)\ub97c \ud1b5\ud574 \uc774\ub8e8\uc5b4\uc9d1\ub2c8\ub2e4.\n<\/p>\n
\n \uc77c\ubc18\uc801\uc73c\ub85c \uc2e0\uacbd\ub9dd\uc740 \uc785\ub825\uce35(Input Layer), \uc740\ub2c9\uce35(Hidden Layers), \ucd9c\ub825\uce35(Output Layer)\uc73c\ub85c \uad6c\uc131\ub429\ub2c8\ub2e4.\n<\/p>\n
\n \ud65c\uc131\ud654 \ud568\uc218\ub294 \ub178\ub4dc\uc5d0\uc11c \ube44\uc120\ud615\uc131\uc744 \ubd80\uc5ec\ud558\ub294 \uc5ed\ud560\uc744 \ud569\ub2c8\ub2e4. \ub2e4\uc74c\uc740 \uc77c\ubc18\uc801\uc73c\ub85c \uc0ac\uc6a9\ub418\ub294 \ud65c\uc131\ud654 \ud568\uc218\uc785\ub2c8\ub2e4.\n<\/p>\n
\n \ub525\ub7ec\ub2dd \ubaa8\ub378\uc744 \ud6c8\ub828\uc2dc\ud0a4\uae30 \uc704\ud574\uc11c\ub294 \ub370\uc774\ud130\uc14b\uc774 \ud544\uc694\ud569\ub2c8\ub2e4. \ub370\uc774\ud130\ub294 \uc785\ub825\uacfc \ubaa9\ud45c(output)\ub85c \uad6c\uc131\ub418\uc5b4 \uc788\uc2b5\ub2c8\ub2e4.
\n \ubaa8\ub378\uc758 \ud559\uc2b5\uc740 \ub2e4\uc74c\uacfc \uac19\uc740 \uc808\ucc28\ub85c \uc9c4\ud589\ub429\ub2c8\ub2e4.\n<\/p>\n
\n \ubaa8\ub378\uc5d0 \uc785\ub825 \ub370\uc774\ud130\ub97c \ud1b5\uacfc\uc2dc\ud0a4\uba70 \uc608\uce21 \uac12\uc744 \uacc4\uc0b0\ud569\ub2c8\ub2e4. \uc774 \ub54c \uc2e0\uacbd\ub9dd\uc758 \uac00\uc911\uce58\uc640 \ud3b8\ud5a5\uc744 \uc0ac\uc6a9\ud558\uc5ec \ucd9c\ub825\uc744 \uc0dd\uc131\ud569\ub2c8\ub2e4.\n<\/p>\n
\n \ubaa8\ub378\uc758 \uc608\uce21 \uacb0\uacfc\uc640 \uc2e4\uc81c \ubaa9\ud45c\uac12 \uac04\uc758 \ucc28\uc774\ub97c \uacc4\uc0b0\ud558\uc5ec \uc190\uc2e4(loss)\uc744 \uad6c\ud569\ub2c8\ub2e4. \uc77c\ubc18\uc801\uc778 \uc190\uc2e4 \ud568\uc218\uc5d0\ub294 \ud3c9\uade0 \uc81c\uacf1 \uc624\ucc28(MSE), \ud06c\ub85c\uc2a4 \uc5d4\ud2b8\ub85c\ud53c \ub4f1\uc774 \uc788\uc2b5\ub2c8\ub2e4.\n<\/p>\n
\n \uc190\uc2e4\uc5d0 \uae30\ubc18\ud558\uc5ec \uac00\uc911\uce58\uc640 \ud3b8\ud5a5\uc744 \uc870\uc815\ud558\ub294 \uacfc\uc815\uc73c\ub85c, \uacbd\uc0ac\ud558\uac15\ubc95(Gradient Descent)\uc744 \uc0ac\uc6a9\ud558\uc5ec \ubaa8\ub378\uc758 \ud30c\ub77c\ubbf8\ud130\ub97c \uc5c5\ub370\uc774\ud2b8\ud569\ub2c8\ub2e4. \uc5ed\uc804\ud30c \uc54c\uace0\ub9ac\uc998\uc740 \uccb4\uc778 \ub8f0\uc744 \uc774\uc6a9\ud558\uc5ec \uac01 \uac00\uc911\uce58\uc5d0 \ub300\ud55c \uc190\uc2e4\uc758 \uae30\uc6b8\uae30\ub97c \uacc4\uc0b0\ud569\ub2c8\ub2e4.\n<\/p>\n
\n \uacc4\uc0b0\ub41c \uae30\uc6b8\uae30\ub97c \uc0ac\uc6a9\ud558\uc5ec \uac00\uc911\uce58\uc640 \ud3b8\ud5a5\uc744 \uc5c5\ub370\uc774\ud2b8\ud569\ub2c8\ub2e4. \uc5c5\ub370\uc774\ud2b8 \ubc29\uc2dd\uc740 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4.\n<\/p>\n
\nw = w - learning_rate * gradient\nb = b - learning_rate * gradient\n<\/pre>\n\ud30c\uc774\ud1a0\uce58\uc5d0\uc11c\uc758 \uad6c\ud604<\/h2>\n
\n \uc774\uc81c \uc704\uc5d0\uc11c \uc124\uba85\ud55c \ub0b4\uc6a9\uc744 \ubc14\ud0d5\uc73c\ub85c \ud30c\uc774\ud1a0\uce58\uc5d0\uc11c \uac04\ub2e8\ud55c \ub525\ub7ec\ub2dd \ubaa8\ub378\uc744 \uad6c\ud604\ud574 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc774\ubc88 \uc608\uc81c\ub294 MNIST \uc190\uae00\uc528 \uc22b\uc790 \uc778\uc2dd \ub370\uc774\ud130\uc14b\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc190\uae00\uc528 \uc22b\uc790\ub97c \ubd84\ub958\ud558\ub294 \ubaa8\ub378\uc785\ub2c8\ub2e4.\n<\/p>\n
\ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac \uc124\uce58 \ubc0f \uc784\ud3ec\ud2b8<\/h3>\n
\npip install torch torchvision\n<\/pre>\n\nimport torch\nimport torch.nn as nn\nimport torch.optim as optim\nfrom torchvision import datasets, transforms\nfrom torch.utils.data import DataLoader\n<\/pre>\n\ub370\uc774\ud130\uc14b \ub85c\ub4dc \ubc0f \uc804\ucc98\ub9ac<\/h3>\n
\n MNIST \ub370\uc774\ud130\uc14b\uc744 \ub85c\ub4dc\ud558\uace0, \uc774\ubbf8\uc9c0 \ub370\uc774\ud130\uc5d0 \ub300\ud574 \uc815\uaddc\ud654\ub97c \uc218\ud589\ud569\ub2c8\ub2e4.\n<\/p>\n
\n# \ub370\uc774\ud130 \uc804\ucc98\ub9ac\ntransform = transforms.Compose([\n transforms.ToTensor(),\n transforms.Normalize((0.5,), (0.5,))\n])\n\n# \ub370\uc774\ud130\uc14b \ub85c\ub4dc\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\n# \ub370\uc774\ud130 \ub85c\ub354 \uc0dd\uc131\ntrain_loader = DataLoader(dataset=train_dataset, batch_size=64, shuffle=True)\ntest_loader = DataLoader(dataset=test_dataset, batch_size=64, shuffle=False)\n<\/pre>\n\ubaa8\ub378 \uc815\uc758<\/h3>\n
\n \uac04\ub2e8\ud55c \uc2e0\uacbd\ub9dd \ubaa8\ub378\uc744 \uc815\uc758\ud569\ub2c8\ub2e4. \uc785\ub825 \ud06c\uae30\ub294 28×28 (MNIST \uc774\ubbf8\uc9c0 \ud06c\uae30)\uc774\uba70, \ub450 \uac1c\uc758 \uc740\ub2c9\uce35\uc744 \uac00\uc9d1\ub2c8\ub2e4. \ucd9c\ub825\uce35\uc740 10 (0~9 \uc22b\uc790)\ub85c \uc124\uc815\ud569\ub2c8\ub2e4.\n<\/p>\n
\nclass SimpleNN(nn.Module):\n def __init__(self):\n super(SimpleNN, self).__init__()\n self.fc1 = nn.Linear(28 * 28, 128) # \uc785\ub825\uce35 -> \uc740\ub2c9\uce35\n self.fc2 = nn.Linear(128, 64) # \uc740\ub2c9\uce35 -> \uc740\ub2c9\uce35\n self.fc3 = nn.Linear(64, 10) # \uc740\ub2c9\uce35 -> \ucd9c\ub825\uce35\n self.activation = nn.ReLU() # \ud65c\uc131\ud654 \ud568\uc218\n\n def forward(self, x):\n x = x.view(-1, 28 * 28) # \uc774\ubbf8\uc9c0\ub97c 1D \ud150\uc11c\ub85c \ubcc0\ud658\n x = self.activation(self.fc1(x)) # \uc804\ubc29 \ud328\uc2a4\n x = self.activation(self.fc2(x))\n x = self.fc3(x)\n return x\n<\/pre>\n\ubaa8\ub378 \ucd08\uae30\ud654 \ubc0f \uc190\uc2e4 \ud568\uc218, \ucd5c\uc801\ud654 \uae30\ubc95 \uc124\uc815<\/h3>\n
\n# \ubaa8\ub378 \ucd08\uae30\ud654\nmodel = SimpleNN()\n\n# \uc190\uc2e4 \ud568\uc218\uc640 \ucd5c\uc801\ud654 \uae30\ubc95 \uc124\uc815\ncriterion = nn.CrossEntropyLoss()\noptimizer = optim.Adam(model.parameters(), lr=0.001)\n<\/pre>\n\ubaa8\ub378 \ud6c8\ub828<\/h3>\n
\n \ubaa8\ub378\uc744 \ubc18\ubcf5\uc801\uc73c\ub85c \ud6c8\ub828\uc2dc\ud0a4\uba74\uc11c \uc190\uc2e4\uc744 \uae30\ub85d\ud558\uace0, \uc8fc\uae30\uc801\uc73c\ub85c \ubaa8\ub378\uc758 \uc131\ub2a5\uc744 \ud3c9\uac00\ud569\ub2c8\ub2e4.\n<\/p>\n
\n# \ubaa8\ub378 \ud6c8\ub828\nnum_epochs = 5\n\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) # \uc804\ubc29 \ud328\uc2a4\n loss = criterion(outputs, labels) # \uc190\uc2e4 \uacc4\uc0b0\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<\/pre>\n\ubaa8\ub378 \ud3c9\uac00<\/h3>\n
\n \ud14c\uc2a4\ud2b8 \ub370\uc774\ud130\uc14b\uc744 \uc0ac\uc6a9\ud558\uc5ec \ubaa8\ub378\uc758 \uc815\ud655\ub3c4\ub97c \ud3c9\uac00\ud569\ub2c8\ub2e4.\n<\/p>\n
\n# \ubaa8\ub378 \ud3c9\uac00\nmodel.eval() # \ud3c9\uac00 \ubaa8\ub4dc\ub85c \uc124\uc815\nwith torch.no_grad(): # \uae30\uc6b8\uae30 \uacc4\uc0b0 \ube44\ud65c\uc131\ud654\n correct = 0\n total = 0\n for images, labels in test_loader:\n outputs = model(images)\n _, predicted = torch.max(outputs.data, 1) # \uc608\uce21\ub41c \ud074\ub798\uc2a4\n total += labels.size(0) # \uc804\uccb4 \uc0d8\ud50c \uc218\n correct += (predicted == labels).sum().item() # \uc815\ub2f5 \uc218\n\nprint(f'Accuracy of the model on the test images: {100 * correct \/ total:.2f}%')\n<\/pre>\n\uacb0\ub860<\/h2>\n
\n \ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 \ub525\ub7ec\ub2dd\uc758 \uae30\ubcf8 \uac1c\ub150\uacfc \ud30c\uc774\ud1a0\uce58\ub97c \ud65c\uc6a9\ud55c \uac04\ub2e8\ud55c \uc2e0\uacbd\ub9dd \ubaa8\ub378 \uad6c\ud604\uc744 \ub2e4\ub918\uc2b5\ub2c8\ub2e4. \uc2e4\uc2b5\uc744 \ud1b5\ud574 \ub370\uc774\ud130 \uc804\ucc98\ub9ac, \ubaa8\ub378 \uc815\uc758, \ud6c8\ub828 \ubc0f \ud3c9\uac00 \uacfc\uc815\uc744 \ubc30\uc6b8 \uc218 \uc788\uc5c8\uc2b5\ub2c8\ub2e4.
\n \uc774\ub97c \ud1b5\ud574 \ub525\ub7ec\ub2dd\uc758 \uc791\ub3d9 \uc6d0\ub9ac\uc5d0 \ub300\ud574 \uae4a\uc774 \uc774\ud574\ud560 \uc218 \uc788\ub294 \uae30\ud68c\ub97c \uc81c\uacf5\ud558\uc600\uc2b5\ub2c8\ub2e4.
\n \ub354 \ub098\uc544\uac00 \ubcf5\uc7a1\ud55c \uc544\ud0a4\ud14d\ucc98, \uace0\uae09 \ucd5c\uc801\ud654 \uae30\ubc95 \ubc0f \ub2e4\uc591\ud55c \ub370\uc774\ud130\uc14b\uc744 \ub2e4\ub8e8\uc5b4 \ub525\ub7ec\ub2dd\uc758 \uc138\uacc4\ub97c \ud0d0\uad6c\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n<\/p>\n\uac10\uc0ac\ud569\ub2c8\ub2e4! <\/p>\n
<\/body><\/p>\n","protected":false},"excerpt":{"rendered":"