\ub525\ub7ec\ub2dd\uc740 \ucd5c\uadfc \uba87 \ub144\uac04 \ub180\ub77c\uc6b4 \ubc1c\uc804\uc744 \ubcf4\uc5ec\uc8fc\uba70 \ub2e4\uc591\ud55c \ubd84\uc57c\uc5d0\uc11c \ud070 \uc601\ud5a5\uc744 \ubbf8\uce58\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uadf8\uc911\uc5d0\uc11c\ub3c4 \uc0dd\uc131 \ubaa8\ub378\uc740 \ub370\uc774\ud130 \uc0d8\ud50c\uc744 \uc0dd\uc131\ud558\ub294 \ub2a5\ub825 \ub355\ubd84\uc5d0 \uc8fc\ubaa9\ubc1b\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc774 \uae00\uc5d0\uc11c\ub294 \uc0dd\uc131 \ubaa8\ub378\uc758 \uc5ec\ub7ec \uc720\ud615\uc744 \uc0b4\ud3b4\ubcf4\uace0 \uac01 \ubaa8\ub378\uc758 \uc791\ub3d9 \uc6d0\ub9ac\uc640 \ud30c\uc774\ud1a0\uce58\ub97c \ud65c\uc6a9\ud55c \uc608\uc81c \ucf54\ub4dc\ub97c \uc81c\uacf5\ud560 \uac83\uc785\ub2c8\ub2e4.<\/p>\n
\uc0dd\uc131 \ubaa8\ub378\uc740 \uc8fc\uc5b4\uc9c4 \ub370\uc774\ud130 \ubd84\ud3ec\uc5d0\uc11c \uc0c8\ub85c\uc6b4 \uc0d8\ud50c\uc744 \uc0dd\uc131\ud558\ub294 \uba38\uc2e0\ub7ec\ub2dd \ubaa8\ub378\uc785\ub2c8\ub2e4. \uc774\uac83\uc740 \uc8fc\uc5b4\uc9c4 \ub370\uc774\ud130\uc640 \uc720\uc0ac\ud558\uc9c0\ub9cc \uc2e4\uc81c \ub370\uc774\ud130\uc5d0\ub294 \uc5c6\ub294 \uc0c8\ub85c\uc6b4 \ub370\uc774\ud130\ub97c \ub9cc\ub4e4\uc5b4\ub0bc \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc0dd\uc131 \ubaa8\ub378\uc740 \uc8fc\ub85c \uc774\ubbf8\uc9c0 \uc0dd\uc131, \ud14d\uc2a4\ud2b8 \uc0dd\uc131, \uc74c\uc545 \uc0dd\uc131 \ub4f1 \ub2e4\uc591\ud55c \ubd84\uc57c\uc5d0 \uc0ac\uc6a9\ub429\ub2c8\ub2e4. \uc0dd\uc131 \ubaa8\ub378\uc758 \uc8fc\uc694 \uc720\ud615\uc740 \ub2e4\uc74c\uacfc \uac19\uc2b5\ub2c8\ub2e4:<\/p>\n
\uc624\ud1a0\uc778\ucf54\ub354\ub294 \uc785\ub825 \ub370\uc774\ud130\ub97c \uc555\ucd95\ud558\uace0 \uc555\ucd95\ub41c \ud45c\ud604\uc73c\ub85c\ubd80\ud130 \uc785\ub825 \ub370\uc774\ud130\ub97c \uc7ac\uad6c\uc131\ud558\ub294 \ubc29\uc2dd\uc73c\ub85c \uc791\ub3d9\ud558\ub294 \uc778\uacf5 \uc2e0\uacbd\ub9dd\uc785\ub2c8\ub2e4. \uc624\ud1a0\uc778\ucf54\ub354\ub294 \uc7a0\uc7ac \uacf5\uac04(latent space)\uc744 \ud1b5\ud574 \ub370\uc774\ud130\ub97c \uc0dd\uc131\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\uc624\ud1a0\uc778\ucf54\ub354\ub294 \ud06c\uac8c \ub450 \ubd80\ubd84\uc73c\ub85c \ub098\ub258\uc5b4 \uc788\uc2b5\ub2c8\ub2e4:<\/p>\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\n# \ub370\uc774\ud130 \uc804\ucc98\ub9ac\ntransform = transforms.Compose([\n transforms.ToTensor(),\n transforms.Normalize((0.5,), (0.5,))\n])\n\n# MNIST \ub370\uc774\ud130\uc14b \ub85c\ub4dc\ntrain_dataset = datasets.MNIST(root='.\/data', train=True, transform=transform, download=True)\ntrain_loader = DataLoader(dataset=train_dataset, batch_size=64, shuffle=True)\n\n# \uc624\ud1a0\uc778\ucf54\ub354 \ubaa8\ub378 \uc815\uc758\nclass Autoencoder(nn.Module):\n def __init__(self):\n super(Autoencoder, self).__init__()\n self.encoder = nn.Sequential(\n nn.Linear(784, 256),\n nn.ReLU(),\n nn.Linear(256, 64)\n )\n self.decoder = nn.Sequential(\n nn.Linear(64, 256),\n nn.ReLU(),\n nn.Linear(256, 784),\n nn.Sigmoid()\n )\n\n def forward(self, x):\n x = x.view(-1, 784) # 28*28 = 784\n encoded = self.encoder(x)\n decoded = self.decoder(encoded)\n return decoded\n\n# \ubaa8\ub378, \uc190\uc2e4 \ud568\uc218, \ucd5c\uc801\ud654 \uc54c\uace0\ub9ac\uc998 \uc815\uc758\nmodel = Autoencoder()\ncriterion = nn.BCELoss()\noptimizer = optim.Adam(model.parameters(), lr=0.001)\n\n# \ud559\uc2b5\nnum_epochs = 10\nfor epoch in range(num_epochs):\n for data in train_loader:\n img, _ = data\n optimizer.zero_grad()\n output = model(img)\n loss = criterion(output, img.view(-1, 784))\n loss.backward()\n optimizer.step()\n print(f'Epoch [{epoch+1}\/{num_epochs}], Loss: {loss.item():.4f}')\n <\/pre>\n\uc704\uc758 \ucf54\ub4dc\ub294 \uc624\ud1a0\uc778\ucf54\ub354\ub97c \uc0ac\uc6a9\ud558\uc5ec MNIST \ub370\uc774\ud130\ub97c \ud559\uc2b5\ud558\ub294 \uac04\ub2e8\ud55c \uc608\uc81c\uc785\ub2c8\ub2e4. \uc778\ucf54\ub354\ub294 784\uac1c\uc758 \uc785\ub825 \ub178\ub4dc\ub97c 64\uac1c\uc758 \uc7a0\uc7ac \ubcc0\uc218\ub85c \ucd95\uc18c\ud558\uace0 \ub514\ucf54\ub354\ub294 \uc774\ub97c \ub2e4\uc2dc 784\uac1c\uc758 \ucd9c\ub825\uc73c\ub85c \ubcf5\uc6d0\ud569\ub2c8\ub2e4.<\/p>\n
2. \uc0dd\uc131\uc801 \uc801\ub300 \uc2e0\uacbd\ub9dd (GANs)<\/h2>\n
GANs\ub294 \ub450 \uac1c\uc758 \uc2e0\uacbd\ub9dd\uc778 \uc0dd\uc131\uae30(Generator)\uc640 \ud310\ubcc4\uae30(Discriminator)\uac00 \uacbd\uc7c1\uc801\uc73c\ub85c \ud559\uc2b5\ud558\ub294 \uad6c\uc870\uc785\ub2c8\ub2e4. \uc0dd\uc131\uae30\ub294 \uc2e4\uc81c \ub370\uc774\ud130\uc640 \uc720\uc0ac\ud55c \uac00\uc9dc \ub370\uc774\ud130\ub97c \uc0dd\uc131\ud558\uace0, \ud310\ubcc4\uae30\ub294 \ub370\uc774\ud130\uac00 \uc2e4\uc81c\uc778\uc9c0 \uac00\uc9dc\uc778\uc9c0 \ud310\ubcc4\ud569\ub2c8\ub2e4.<\/p>\n
GANs\uc758 \uc791\ub3d9 \uc6d0\ub9ac<\/h3>\n
GANs\uc758 \ud559\uc2b5 \uacfc\uc815\uc740 \ub2e4\uc74c\uacfc \uac19\uc740 \ubc29\uc2dd\uc73c\ub85c \uc9c4\ud589\ub429\ub2c8\ub2e4:<\/p>\n
\n- \uc0dd\uc131\uae30\ub294 \ubb34\uc791\uc704 \ub178\uc774\uc988\ub97c \uc785\ub825\uc73c\ub85c \ubc1b\uc544 \uac00\uc9dc \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud569\ub2c8\ub2e4.<\/li>\n
- \ud310\ubcc4\uae30\ub294 \uc2e4\uc81c \uc774\ubbf8\uc9c0\uc640 \uc0dd\uc131\ub41c \uc774\ubbf8\uc9c0\ub97c \uc785\ub825\uc73c\ub85c \ubc1b\uc544 \ub450 \uac00\uc9c0 \uc774\ubbf8\uc9c0\uc758 \uc9c4\uc704\ub97c \ud310\ub2e8\ud569\ub2c8\ub2e4.<\/li>\n
- \ud310\ubcc4\uae30\uac00 \uac00\uc9dc \uc774\ubbf8\uc9c0\ub97c \uc798 \ud310\ubcc4\ud560\uc218\ub85d \uc0dd\uc131\uae30\ub294 \ubcf4\ub2e4 \ub354 \uc815\uad50\ud55c \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\uae30 \uc704\ud574 \ud559\uc2b5\ud569\ub2c8\ub2e4.<\/li>\n<\/ol>\n
\ud30c\uc774\ud1a0\uce58\ub85c GAN \ubaa8\ub378 \ub9cc\ub4e4\uae30<\/h3>\n\nclass Generator(nn.Module):\n def __init__(self):\n super(Generator, self).__init__()\n self.model = nn.Sequential(\n nn.Linear(100, 256),\n nn.ReLU(),\n nn.Linear(256, 512),\n nn.ReLU(),\n nn.Linear(512, 1024),\n nn.ReLU(),\n nn.Linear(1024, 784),\n nn.Tanh()\n )\n\n def forward(self, x):\n return self.model(x)\n\nclass Discriminator(nn.Module):\n def __init__(self):\n super(Discriminator, self).__init__()\n self.model = nn.Sequential(\n nn.Linear(784, 512),\n nn.LeakyReLU(0.2),\n nn.Linear(512, 256),\n nn.LeakyReLU(0.2),\n nn.Linear(256, 1),\n nn.Sigmoid()\n )\n\n def forward(self, x):\n return self.model(x)\n\n# \ubaa8\ub378 \uc778\uc2a4\ud134\uc2a4 \uc0dd\uc131\ngenerator = Generator()\ndiscriminator = Discriminator()\n\n# \uc190\uc2e4 \ud568\uc218 \ubc0f \ucd5c\uc801\ud654\uae30 \uc815\uc758\ncriterion = nn.BCELoss()\noptimizer_g = optim.Adam(generator.parameters(), lr=0.0002)\noptimizer_d = optim.Adam(discriminator.parameters(), lr=0.0002)\n\n# \ud559\uc2b5 \uacfc\uc815\nnum_epochs = 100\nfor epoch in range(num_epochs):\n for data in train_loader:\n real_images, _ = data\n real_labels = torch.ones(real_images.size(0), 1)\n fake_labels = torch.zeros(real_images.size(0), 1)\n\n # \ud310\ubcc4\uae30 \ud559\uc2b5\n optimizer_d.zero_grad()\n outputs = discriminator(real_images.view(-1, 784))\n d_loss_real = criterion(outputs, real_labels)\n d_loss_real.backward()\n\n noise = torch.randn(real_images.size(0), 100)\n fake_images = generator(noise)\n outputs = discriminator(fake_images.detach())\n d_loss_fake = criterion(outputs, fake_labels)\n d_loss_fake.backward()\n\n optimizer_d.step()\n\n # \uc0dd\uc131\uae30 \ud559\uc2b5\n optimizer_g.zero_grad()\n outputs = discriminator(fake_images)\n g_loss = criterion(outputs, real_labels)\n g_loss.backward()\n optimizer_g.step()\n\n print(f'Epoch [{epoch+1}\/{num_epochs}], d_loss: {d_loss_real.item() + d_loss_fake.item():.4f}, g_loss: {g_loss.item():.4f}')\n <\/pre>\n\uc704\uc758 \ucf54\ub4dc\ub294 GAN\uc744 \uad6c\ud604\ud558\ub294 \uae30\ubcf8\uc801\uc778 \uc608\uc81c\uc785\ub2c8\ub2e4. Generator\ub294 100\ucc28\uc6d0\uc758 \ub79c\ub364 \ub178\uc774\uc988\ub97c \uc785\ub825\ubc1b\uc544 784\ucc28\uc6d0\uc758 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\uace0, Discriminator\ub294 \uc774\ub7ec\ud55c \uc774\ubbf8\uc9c0\ub97c \ud310\ubcc4\ud569\ub2c8\ub2e4.<\/p>\n
3. \ubcc0\ubd84 \uc624\ud1a0\uc778\ucf54\ub354 (Variational Autoencoders, VAEs)<\/h2>\n
VAEs\ub294 \uc624\ud1a0\uc778\ucf54\ub354\uc758 \ud655\uc7a5\uc73c\ub85c, \uc0dd\uc131 \ubaa8\ub378\uc785\ub2c8\ub2e4. VAEs\ub294 \ub370\uc774\ud130\uc758 \uc7a0\uc7ac \ubd84\ud3ec\ub97c \ud559\uc2b5\ud558\uc5ec \uc0c8\ub85c\uc6b4 \uc0d8\ud50c\uc744 \uc0dd\uc131\ud560 \uc218 \uc788\ub294 \ubaa8\ub378\uc785\ub2c8\ub2e4. \uc11c\ub85c \ub2e4\ub978 \ub370\uc774\ud130 \ud3ec\uc778\ud2b8\uc758 \uc7a0\uc7ac \ubcc0\uc218\ub97c \uc0d8\ud50c\ub9c1\ud558\uc5ec \ub2e4\uc591\ud55c \uc0d8\ud50c\uc744 \uc0dd\uc131 \ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
VAE\uc758 \uad6c\uc870<\/h3>\n
VAE\ub294 \ubcc0\ubd84 \ucd94\uc815 \uae30\ubc95\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc785\ub825 \ub370\uc774\ud130\ub97c \uc7a0\uc7ac \uacf5\uac04\uc5d0 \ub9e4\ud551\ud569\ub2c8\ub2e4. VAE\ub294 \uc778\ucf54\ub354\uc640 \ub514\ucf54\ub354\ub85c \uad6c\uc131\ub418\uc5b4 \uc788\uc73c\uba70, \uc778\ucf54\ub354\ub294 \uc785\ub825 \ub370\uc774\ud130\ub97c \ud3c9\uade0\uacfc \ubd84\uc0b0\uc73c\ub85c \ub9e4\ud551\ud558\uace0, \uc0d8\ud50c\ub9c1 \uacfc\uc815\uc744 \ud1b5\ud574 \ub370\uc774\ud130 \ud3ec\uc778\ud2b8\ub97c \uc0dd\uc131\ud569\ub2c8\ub2e4.<\/p>\n
\ud30c\uc774\ud1a0\uce58\ub85c VAE \ubaa8\ub378 \ub9cc\ub4e4\uae30<\/h3>\n\nclass VAE(nn.Module):\n def __init__(self):\n super(VAE, self).__init__()\n self.encoder = nn.Sequential(\n nn.Linear(784, 256),\n nn.ReLU(),\n nn.Linear(256, 128),\n nn.ReLU()\n )\n self.fc_mean = nn.Linear(128, 20)\n self.fc_logvar = nn.Linear(128, 20)\n self.decoder = nn.Sequential(\n nn.Linear(20, 128),\n nn.ReLU(),\n nn.Linear(128, 256),\n nn.ReLU(),\n nn.Linear(256, 784),\n nn.Sigmoid()\n )\n\n def encode(self, x):\n h = self.encoder(x.view(-1, 784))\n return self.fc_mean(h), self.fc_logvar(h)\n\n def reparameterize(self, mu, logvar):\n std = torch.exp(0.5 * logvar)\n eps = torch.randn_like(std)\n return mu + eps * std\n\n def decode(self, z):\n return self.decoder(z)\n\n def forward(self, x):\n mu, logvar = self.encode(x)\n z = self.reparameterize(mu, logvar)\n return self.decode(z), mu, logvar\n\n# \uc190\uc2e4 \ud568\uc218 \uc815\uc758\ndef loss_function(recon_x, x, mu, logvar):\n BCE = nn.functional.binary_cross_entropy(recon_x, x.view(-1, 784), reduction='sum')\n KLD = -0.5 * torch.sum(1 + logvar - mu.pow(2) - logvar.exp())\n return BCE + KLD\n\n# \ubaa8\ub378 \ucd08\uae30\ud654 \ubc0f \ud559\uc2b5 \uacfc\uc815\nmodel = VAE()\noptimizer = optim.Adam(model.parameters(), lr=0.001)\n\n# \ud559\uc2b5 \uacfc\uc815\nnum_epochs = 10\nfor epoch in range(num_epochs):\n for data in train_loader:\n img, _ = data\n optimizer.zero_grad()\n recon_batch, mu, logvar = model(img)\n loss = loss_function(recon_batch, img, mu, logvar)\n loss.backward()\n optimizer.step()\n print(f'Epoch [{epoch+1}\/{num_epochs}], Loss: {loss.item():.4f}')\n <\/pre>\n4. \uc5f0\uad6c \ub3d9\ud5a5 \ubc0f \uacb0\ub860<\/h2>\n
\uc0dd\uc131 \ubaa8\ub378\uc740 \uc2e0\ub8b0\uc131\uc774 \ub192\uc740 \ub370\uc774\ud130 \uc0dd\uc131\uc774 \uac00\ub2a5\ud558\uc5ec \uc5ec\ub7ec \ubd84\uc57c\uc5d0\uc11c \uc751\uc6a9\ub418\uace0 \uc788\uc2b5\ub2c8\ub2e4. GANs, VAEs, \uc624\ud1a0\uc778\ucf54\ub354\ub294 \ud2b9\ud788 \uc774\ubbf8\uc9c0 \uc0dd\uc131, \ube44\ub514\uc624 \uc0dd\uc131, \ud14d\uc2a4\ud2b8 \uc0dd\uc131 \ub4f1\uc758 \uc560\ud50c\ub9ac\ucf00\uc774\uc158\uc5d0\uc11c \ud3ed\ub113\uac8c \uc0ac\uc6a9\ub418\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \ubaa8\ub378\ub4e4\uc740 \uae4a\uc774 \uc788\ub294 \ud559\uc2b5\uacfc \ub354\ubd88\uc5b4 \ub370\uc774\ud130 \uacfc\ud559, \uc778\uacf5\uc9c0\ub2a5\uc5d0\uc11c\uc758 \uc0ac\uc6a9 \uac00\ub2a5\uc131\uc744 \uadf9\ub300\ud654\ud558\uace0 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\ub525\ub7ec\ub2dd \uae30\uc220\uc758 \ubc1c\uc804\uacfc \ub354\ubd88\uc5b4 \uc0dd\uc131 \ubaa8\ub378 \ub610\ud55c \uacc4\uc18d\ud574\uc11c \ubc1c\uc804\ud558\uace0 \uc788\uc73c\uba70, \uc774 \uae00\uc5d0\uc11c \ub2e4\ub8ec \uae30\ucd08 \uac1c\ub150\uacfc \uc608\uc81c\ub4e4\uc744 \ubc14\ud0d5\uc73c\ub85c \ub354 \ub2e4\uc591\ud55c \uc2e4\ud5d8\uacfc \uc5f0\uad6c\uac00 \ud544\uc694\ud569\ub2c8\ub2e4.<\/p>\n
\ub525\ub7ec\ub2dd\uc744 \ud1b5\ud55c \uc0dd\uc131 \ubaa8\ub378\uc758 \uc751\uc6a9 \uac00\ub2a5\uc131\uc5d0 \ub300\ud574 \ub354 \uae4a\uc774 \ub098\uc544\uac00\uace0 \uc2f6\ub2e4\uba74 \ub17c\ubb38\uc774\ub098 \uc2ec\ud654 \ud559\uc2b5 \uc790\ub8cc\ub97c \ucc38\uace0\ud558\uc5ec \ub354 \ub9ce\uc740 \uc0ac\ub840\ub97c \uc5f0\uad6c\ud574 \ubcf4\ub294 \uac83\uc744 \ucd94\ucc9c\ub4dc\ub9bd\ub2c8\ub2e4.<\/p>\n
\uc774 \ud3ec\uc2a4\ud2b8\uac00 \uc0dd\uc131 \ubaa8\ub378\uc5d0 \ub300\ud55c \uc774\ud574\ub97c \ub3d5\uace0 \ub525\ub7ec\ub2dd\uc758 \ub9e4\ub825\uc744 \ub290\ub07c\ub294 \ub370 \ub3c4\uc6c0\uc774 \ub418\uae30\ub97c \ubc14\ub78d\ub2c8\ub2e4.<\/p>\n
<\/body><\/p>\n","protected":false},"excerpt":{"rendered":"