\ucd5c\uadfc \uc778\uacf5\uc9c0\ub2a5 \ubd84\uc57c\uc5d0\uc11c \uc0dd\uc131\uc801 \uc801\ub300 \uc2e0\uacbd\ub9dd(GAN)\uacfc \ubcc0\ud615 \uc624\ud1a0\uc778\ucf54\ub354(VAE)\ub294 \uc774\ubbf8\uc9c0 \uc0dd\uc131\uc758 \ud6a8\uc728\uacfc \ud488\uc9c8\uc744 \ud06c\uac8c \ud5a5\uc0c1\uc2dc\ud0a4\ub294 \uc911\uc694\ud55c \uae30\uc220\ub85c \uc790\ub9ac \uc7a1\uc558\uc2b5\ub2c8\ub2e4. \uc774 \uae00\uc5d0\uc11c\ub294 GAN\uacfc VAE\uc758 \uae30\ubcf8 \uac1c\ub150\uacfc \ud568\uaed8, \ud30c\uc774\ud1a0\uce58\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc5bc\uad74 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\ub294 \uacfc\uc815\uc744 \uc790\uc138\ud788 \uc0b4\ud3b4\ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
1. GAN(Generative Adversarial Networks) \uac1c\uc694<\/h2>\n
Generative Adversarial Networks(GAN)\ub294 \ub450 \uac1c\uc758 \uc2e0\uacbd\ub9dd\uc778 \uc0dd\uc131\uc790(Generator)\uc640 \ud310\ubcc4\uc790(Discriminator)\uac00 \uc11c\ub85c \uacbd\uc7c1\ud558\uba70 \ud559\uc2b5\ud558\ub294 \uad6c\uc870\ub97c \uac00\uc9c0\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc0dd\uc131\uc790\ub294 \uc2e4\uc81c\uc640 \uc720\uc0ac\ud55c \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\ub824\uace0 \ud558\uace0, \ud310\ubcc4\uc790\ub294 \uc0dd\uc131\ub41c \uc774\ubbf8\uc9c0\uac00 \uc9c4\uc9dc\uc778\uc9c0 \uac00\uc9dc\uc778\uc9c0\ub97c \ud310\ubcc4\ud569\ub2c8\ub2e4. \uc774 \uacfc\uc815\uc740 \uc0dd\uc131\uc790\uac00 \ud310\ubcc4\uc790\ub97c \uc18d\uc774\ub3c4\ub85d \ud559\uc2b5\ud558\uba74\uc11c \uc810\uc810 \ub354 \ud604\uc2e4\uc801\uc778 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\ub294 \ub370 \ub3c4\uc6c0\uc744 \uc8fc\uac8c \ub429\ub2c8\ub2e4.<\/p>\n
1.1 GAN\uc758 \ub3d9\uc791 \uc6d0\ub9ac<\/h3>\n
GAN\uc740 \ub2e4\uc74c\uacfc \uac19\uc740 \ub450 \uac00\uc9c0\uc758 \ub124\ud2b8\uc6cc\ud06c\ub85c \uad6c\uc131\ub429\ub2c8\ub2e4:<\/p>\n
- \n
- \uc0dd\uc131\uc790(Generator):<\/strong> \ub79c\ub364 \ub178\uc774\uc988\ub97c \uc785\ub825\uc73c\ub85c \ubc1b\uc544 \uc2e4\uc81c\uc640 \uc720\uc0ac\ud55c \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud569\ub2c8\ub2e4.<\/li>\n
- \ud310\ubcc4\uc790(Discriminator):<\/strong> \uc785\ub825\ub41c \uc774\ubbf8\uc9c0\uac00 \uc9c4\uc9dc\uc778\uc9c0 \uac00\uc9dc\uc778\uc9c0 \ubd84\ub958\ud569\ub2c8\ub2e4.<\/li>\n<\/ul>\n
\ud559\uc2b5\uc774 \uc9c4\ud589\ub428\uc5d0 \ub530\ub77c \uc0dd\uc131\uc790\ub294 \uc810\ucc28 \ud488\uc9c8 \ub192\uc740 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\uace0, \ud310\ubcc4\uc790\ub294 \ub354\uc6b1 \uc815\ud655\ud558\uac8c \uc774\ubbf8\uc9c0\ub97c \ubd84\uc11d\ud569\ub2c8\ub2e4. \uc774 \uacfc\uc815\uc740 \uc81c\ub85c\uc12c \uac8c\uc784\uc758 \ud615\ud0dc\ub85c \uc9c4\ud589\ub418\uba70, GAN \ubaa8\ub378\uc758 \ubaa9\ud45c\ub294 \uc774 \ub450 \ub124\ud2b8\uc6cc\ud06c\uc758 \uc131\ub2a5\uc744 \ub3d9\uc2dc\uc5d0 \ud5a5\uc0c1\uc2dc\ud0a4\ub294 \uac83\uc785\ub2c8\ub2e4.<\/p>\n
2. VAE(Variational Autoencoder) \uac1c\uc694<\/h2>\n
\ubcc0\ud615 \uc624\ud1a0\uc778\ucf54\ub354(VAE)\ub294 \uc774\ubbf8\uc9c0\ub098 \ub370\uc774\ud130\uc758 \uc7a0\uc7ac \uacf5\uac04(latent space)\uc744 \ud559\uc2b5\ud558\uc5ec \uc0c8\ub85c\uc6b4 \ub370\uc774\ud130\ub97c \uc0dd\uc131\ud560 \uc218 \uc788\ub3c4\ub85d \ud558\ub294 \ubaa8\ub378\uc785\ub2c8\ub2e4. VAE\ub294 \uc785\ub825 \ub370\uc774\ud130\ub97c \uc778\ucf54\ub354\ub97c \ud1b5\ud574 \uc800\ucc28\uc6d0 \uc7a0\uc7ac \uacf5\uac04\uc73c\ub85c \ubcc0\ud658\ud55c \ud6c4, \uc774 \uc7a0\uc7ac \uacf5\uac04\uc5d0\uc11c \uc0d8\ud50c\ub9c1\ud558\uc5ec \ub514\ucf54\ub354\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc774\ubbf8\uc9c0\ub97c \uc7ac\uad6c\uc131\ud569\ub2c8\ub2e4. VAE\ub294 \ud655\ub960\uc801 \ubaa8\ub378\ub85c, \uc785\ub825 \ub370\uc774\ud130\uc758 \ubd84\ud3ec\ub97c \ud559\uc2b5\ud558\uace0 \uc774\ub97c \uae30\ubc18\uc73c\ub85c \uc0c8\ub85c\uc6b4 \uc0d8\ud50c\uc744 \uc0dd\uc131\ud569\ub2c8\ub2e4.<\/p>\n
2.1 VAE\uc758 \uad6c\uc870<\/h3>\n
VAE\ub294 \ub2e4\uc74c\uacfc \uac19\uc740 \uc138 \uac00\uc9c0 \uc8fc\uc694 \uad6c\uc131 \uc694\uc18c\ub85c \uc774\ub8e8\uc5b4\uc9d1\ub2c8\ub2e4:<\/p>\n
- \n
- \uc778\ucf54\ub354(Encoder):<\/strong> \uc785\ub825 \ub370\uc774\ud130\ub97c \uc7a0\uc7ac \ubcc0\uc218\ub85c \ubcc0\ud658\ud569\ub2c8\ub2e4.<\/li>\n
- \uc0d8\ud50c\ub9c1(Sampling):<\/strong> \uc7a0\uc7ac \ubcc0\uc218\uc5d0\uc11c \uc0d8\ud50c\uc744 \ucd94\ucd9c\ud569\ub2c8\ub2e4.<\/li>\n
- \ub514\ucf54\ub354(Decoder):<\/strong> \uc0d8\ud50c\ub41c \uc7a0\uc7ac \ubcc0\uc218\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc0c8\ub85c\uc6b4 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud569\ub2c8\ub2e4.<\/li>\n<\/ul>\n
3. \ud504\ub85c\uc81d\ud2b8 \ubaa9\ud45c\uc640 \ub370\uc774\ud130\uc14b<\/h2>\n
\uc774\ubc88 \ud504\ub85c\uc81d\ud2b8\uc758 \ubaa9\ud45c\ub294 GAN \ubc0f VAE\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc2e4\uc81c\uc640 \uc720\uc0ac\ud55c \uc5bc\uad74 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\ub294 \uac83\uc785\ub2c8\ub2e4. \uc774\ub97c \uc704\ud574 CelebA \ub370\uc774\ud130\uc14b\uc744 \uc0ac\uc6a9\ud558\uaca0\uc2b5\ub2c8\ub2e4. CelebA \ub370\uc774\ud130\uc14b\uc5d0\ub294 \ub2e4\uc591\ud55c \uc5bc\uad74 \uc774\ubbf8\uc9c0\uac00 \ud3ec\ud568\ub418\uc5b4 \uc788\uc73c\uba70, GAN\uacfc VAE\uc758 \uc131\ub2a5\uc744 \uce21\uc815\ud558\ub294 \ub370 \uc801\ud569\ud55c \ub370\uc774\ud130\uc14b\uc785\ub2c8\ub2e4.<\/p>\n
4. \ud658\uacbd \uc124\uc815<\/h2>\n
\uc774 \ud504\ub85c\uc81d\ud2b8\ub97c \uc9c4\ud589\ud558\uae30 \uc704\ud574 Python\uacfc PyTorch \ud504\ub808\uc784\uc6cc\ud06c\uac00 \ud544\uc694\ud569\ub2c8\ub2e4. \ub2e4\uc74c\uc740 \ud544\uc694\ud55c \ud328\ud0a4\uc9c0 \ubaa9\ub85d\uc785\ub2c8\ub2e4:<\/p>\n
pip install torch torchvision matplotlib<\/code><\/pre>\n5. \ud30c\uc774\ud1a0\uce58\ub85c GAN \uad6c\ud604\ud558\uae30<\/h2>\n
\uba3c\uc800 GAN \ubaa8\ub378\uc744 \uad6c\ud604\ud574 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4. GAN\uc758 \uad6c\uc870\ub294 \ub2e4\uc74c\uacfc \uac19\uc740 \ub2e8\uacc4\ub85c \uad6c\uc131\ub429\ub2c8\ub2e4:<\/p>\n
- \n
- \ub370\uc774\ud130\uc14b \ub85c\ub529<\/li>\n
- \uc0dd\uc131\uc790\uc640 \ud310\ubcc4\uc790 \uc815\uc758<\/li>\n
- \ud6c8\ub828 \ub8e8\ud504 \uc124\uc815<\/li>\n
- \uacb0\uacfc \uc2dc\uac01\ud654<\/li>\n<\/ul>\n
5.1 \ub370\uc774\ud130\uc14b \ub85c\ub529<\/h3>\n
\uc6b0\uc120, CelebA \ub370\uc774\ud130\uc14b\uc744 \ub2e4\uc6b4\ub85c\ub4dc\ud558\uace0 \uc900\ube44\ud569\ub2c8\ub2e4.<\/p>\n
import torchvision.transforms as transforms\nfrom torchvision.datasets import ImageFolder\nfrom torch.utils.data import DataLoader\n\ntransform = transforms.Compose([\n transforms.Resize((64, 64)),\n transforms.ToTensor(),\n])\n\ndataset = ImageFolder(root='path_to_celeba', transform=transform)\ndataloader = DataLoader(dataset, batch_size=64, shuffle=True)<\/code><\/pre>\n5.2 \uc0dd\uc131\uc790 \ubc0f \ud310\ubcc4\uc790 \uc815\uc758<\/h3>\n
GAN\uc758 \uc0dd\uc131\uc790 \ubc0f \ud310\ubcc4\uc790\ub97c \uc815\uc758\ud569\ub2c8\ub2e4.<\/p>\n
import torch\nimport torch.nn as nn\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, 3 * 64 * 64),\n nn.Tanh(),\n )\n\n def forward(self, z):\n z = self.model(z)\n return z.view(-1, 3, 64, 64)\n\nclass Discriminator(nn.Module):\n def __init__(self):\n super(Discriminator, self).__init__()\n self.model = nn.Sequential(\n nn.Linear(3 * 64 * 64, 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, img):\n img_flat = img.view(img.size(0), -1)\n return self.model(img_flat)<\/code><\/pre>\n5.3 \ud6c8\ub828 \ub8e8\ud504 \uc124\uc815<\/h3>\n
\uc774\uc81c GAN\uc758 \ud6c8\ub828 \uacfc\uc815\uc744 \uad6c\ud604\ud569\ub2c8\ub2e4.<\/p>\n
import torch.optim as optim\n\ndevice = torch.device(\"cuda\" if torch.cuda.is_available() else \"cpu\")\ngenerator = Generator().to(device)\ndiscriminator = Discriminator().to(device)\n\ncriterion = nn.BCELoss()\ng_optimizer = optim.Adam(generator.parameters(), lr=0.0002, betas=(0.5, 0.999))\nd_optimizer = optim.Adam(discriminator.parameters(), lr=0.0002, betas=(0.5, 0.999))\n\nnum_epochs = 50\nfor epoch in range(num_epochs):\n for i, (imgs, _) in enumerate(dataloader):\n imgs = imgs.to(device)\n batch_size = imgs.size(0)\n\n # \ub808\uc774\ube14 \uc124\uc815\n real_labels = torch.ones(batch_size, 1).to(device)\n fake_labels = torch.zeros(batch_size, 1).to(device)\n\n # \ud310\ubcc4\uc790 \ud6c8\ub828\n d_optimizer.zero_grad()\n outputs = discriminator(imgs)\n d_loss_real = criterion(outputs, real_labels)\n d_loss_real.backward()\n\n z = torch.randn(batch_size, 100).to(device)\n fake_imgs = generator(z)\n outputs = discriminator(fake_imgs.detach())\n d_loss_fake = criterion(outputs, fake_labels)\n d_loss_fake.backward()\n \n d_loss = d_loss_real + d_loss_fake\n d_optimizer.step()\n\n # \uc0dd\uc131\uc790 \ud6c8\ub828\n g_optimizer.zero_grad()\n outputs = discriminator(fake_imgs)\n g_loss = criterion(outputs, real_labels)\n g_loss.backward()\n g_optimizer.step()\n\n print(f'Epoch [{epoch+1}\/{num_epochs}], d_loss: {d_loss.item()}, g_loss: {g_loss.item()}')<\/code><\/pre>\n5.4 \uacb0\uacfc \uc2dc\uac01\ud654<\/h3>\n
\ud6c8\ub828\ub41c \uc0dd\uc131\uc790\uac00 \uc0dd\uc131\ud55c \uc774\ubbf8\uc9c0\ub97c \uc2dc\uac01\ud654\ud569\ub2c8\ub2e4.<\/p>\n
import matplotlib.pyplot as plt\n\nz = torch.randn(64, 100).to(device)\nfake_images = generator(z).detach().cpu()\n\nplt.figure(figsize=(8, 8))\nfor i in range(64):\n plt.subplot(8, 8, i + 1)\n plt.imshow(fake_images[i].permute(1, 2, 0).numpy() * 0.5 + 0.5)\n plt.axis('off')\nplt.show()<\/code><\/pre>\n6. \ud30c\uc774\ud1a0\uce58\ub85c VAE \uad6c\ud604\ud558\uae30<\/h2>\n
\uc774\uc81c VAE\ub97c \uad6c\ud604\ud574 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4. VAE\uc758 \uad6c\uc870\ub294 GAN\uacfc \ube44\uc2b7\ud558\uc9c0\ub9cc, \ud655\ub960\uc801\uc778 \uc811\uadfc \ubc29\uc2dd\uc744 \uc0ac\uc6a9\ud569\ub2c8\ub2e4. VAE\uc758 \uad6c\ud604 \ub2e8\uacc4\ub294 \uc544\ub798\uc640 \uac19\uc2b5\ub2c8\ub2e4:<\/p>\n
- \n
- \ub370\uc774\ud130\uc14b \uc900\ube44<\/li>\n
- \uc778\ucf54\ub354\uc640 \ub514\ucf54\ub354 \uc815\uc758<\/li>\n
- \ud6c8\ub828 \ub8e8\ud504 \uc124\uc815<\/li>\n
- \uacb0\uacfc \uc2dc\uac01\ud654<\/li>\n<\/ul>\n
6.1 \ub370\uc774\ud130\uc14b \uc900\ube44<\/h3>\n
\ub370\uc774\ud130\uc14b\uc740 GAN\uc744 \uc0ac\uc6a9\ud560 \ub54c\uc640 \ub3d9\uc77c\ud558\uac8c \ub85c\ub4dc\ud569\ub2c8\ub2e4.<\/p>\n
6.2 \uc778\ucf54\ub354\uc640 \ub514\ucf54\ub354 \uc815\uc758<\/h3>\n
VAE\uc758 \uc778\ucf54\ub354\uc640 \ub514\ucf54\ub354\ub97c \uc815\uc758\ud569\ub2c8\ub2e4.<\/p>\n
class VAE(nn.Module):\n def __init__(self):\n super(VAE, self).__init__()\n self.encoder = nn.Sequential(\n nn.Conv2d(3, 16, 4, stride=2, padding=1),\n nn.ReLU(),\n nn.Conv2d(16, 32, 4, stride=2, padding=1),\n nn.ReLU(),\n nn.Conv2d(32, 64, 4, stride=2, padding=1),\n nn.ReLU(),\n )\n self.fc_mu = nn.Linear(64 * 8 * 8, 128)\n self.fc_logvar = nn.Linear(64 * 8 * 8, 128)\n self.fc_decode = nn.Linear(128, 64 * 8 * 8)\n self.decoder = nn.Sequential(\n nn.ConvTranspose2d(64, 32, 4, stride=2, padding=1),\n nn.ReLU(),\n nn.ConvTranspose2d(32, 16, 4, stride=2, padding=1),\n nn.ReLU(),\n nn.ConvTranspose2d(16, 3, 4, stride=2, padding=1),\n nn.Sigmoid(),\n )\n\n def encode(self, x):\n h = self.encoder(x)\n h = h.view(h.size(0), -1)\n return self.fc_mu(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 z = self.fc_decode(z).view(-1, 64, 8, 8)\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<\/code><\/pre>\n6.3 \ud6c8\ub828 \ub8e8\ud504 \uc124\uc815<\/h3>\n
VAE\uc758 \ud6c8\ub828 \uacfc\uc815\uc744 \uad6c\ud604\ud569\ub2c8\ub2e4. VAE\ub294 \ub450 \uac00\uc9c0 \uc190\uc2e4\uc744 \uc0ac\uc6a9\ud558\uc5ec \ud6c8\ub828\ub418\uba70, \uc6d0\ubcf8 \uc774\ubbf8\uc9c0\uc640 \ubcf5\uc6d0\ub41c \uc774\ubbf8\uc9c0 \uac04\uc758 \ucc28\uc774(\uc7ac\uad6c\uc131 \uc190\uc2e4)\uc640 \uc7a0\uc7ac \uacf5\uac04\uc758 \ubd84\ud3ec\uc640 \uc815\uaddc \ubd84\ud3ec \uac04\uc758 \ucc28\uc774(\ucfe8\ubc31-\ub77c\uc774\ube14\ub7ec \ubc1c\uc0b0 \uc190\uc2e4)\ub85c \uad6c\uc131\ub429\ub2c8\ub2e4.<\/p>\n
vae = VAE().to(device)\noptimizer = optim.Adam(vae.parameters(), lr=0.0002)\n\nnum_epochs = 50\nfor epoch in range(num_epochs):\n for imgs, _ in dataloader:\n imgs = imgs.to(device)\n\n optimizer.zero_grad()\n reconstructed, mu, logvar = vae(imgs)\n\n re_loss = nn.functional.binary_cross_entropy(reconstructed.view(-1, 3 * 64 * 64), imgs.view(-1, 3 * 64 * 64), reduction='sum')\n kl_loss = -0.5 * torch.sum(1 + logvar - mu.pow(2) - logvar.exp())\n loss = re_loss + kl_loss\n\n loss.backward()\n optimizer.step()\n\n print(f'Epoch [{epoch+1}\/{num_epochs}], Loss: {loss.item()}')<\/code><\/pre>\n6.4 \uacb0\uacfc \uc2dc\uac01\ud654<\/h3>\n
\ud6c8\ub828\ub41c VAE\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc774\ubbf8\uc9c0\ub97c \ubcf5\uc6d0\ud558\uace0 \uc2dc\uac01\ud654\ud569\ub2c8\ub2e4.<\/p>\n
with torch.no_grad():\n z = torch.randn(64, 128).to(device)\n generated_images = vae.decode(z).cpu()\n\nplt.figure(figsize=(8, 8))\nfor i in range(64):\n plt.subplot(8, 8, i + 1)\n plt.imshow(generated_images[i].permute(1, 2, 0).numpy())\n plt.axis('off')\nplt.show()<\/code><\/pre>\n7. \uacb0\ub860<\/h2>\n
\uc774 \uae00\uc5d0\uc11c\ub294 \ud30c\uc774\ud1a0\uce58\ub97c \ud65c\uc6a9\ud558\uc5ec GAN\uacfc VAE\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc5bc\uad74 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\ub294 \ubc29\ubc95\uc744 \uc0b4\ud3b4\ubcf4\uc558\uc2b5\ub2c8\ub2e4. GAN\uc740 \uc0dd\uc131\uc790\uc640 \ud310\ubcc4\uc790\uac00 \uc11c\ub85c \uacbd\uc7c1\ud558\uba74\uc11c \uc810\uc810 \ub354 \ud604\uc2e4\uc801\uc778 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\ub3c4\ub85d \ud559\uc2b5\ud558\ub294 \ubc18\uba74, VAE\ub294 \uc7a0\uc7ac \uacf5\uac04\uc758 \ubd84\ud3ec\ub97c \ud559\uc2b5\ud558\uc5ec \uc0c8\ub85c\uc6b4 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud569\ub2c8\ub2e4. \ub450 \uae30\uc220 \ubaa8\ub450 \uc774\ubbf8\uc9c0 \uc0dd\uc131 \ubd84\uc57c\uc5d0\uc11c \uc911\uc694\ud55c \uc5ed\ud560\uc744 \ud558\uace0 \uc788\uc73c\uba70, \uac01\uae30 \ub2e4\ub978 \ubc29\ubc95\uc73c\ub85c \ub180\ub77c\uc6b4 \uacb0\uacfc\ub97c \ub9cc\ub4e4\uc5b4\ub0bc \uc218 \uc788\uc2b5\ub2c8\ub2e4. <\/p>\n
8. \ucd94\uac00 \ucc38\uace0 \uc790\ub8cc<\/h2>\n
- \uc0d8\ud50c\ub9c1(Sampling):<\/strong> \uc7a0\uc7ac \ubcc0\uc218\uc5d0\uc11c \uc0d8\ud50c\uc744 \ucd94\ucd9c\ud569\ub2c8\ub2e4.<\/li>\n
- \ud310\ubcc4\uc790(Discriminator):<\/strong> \uc785\ub825\ub41c \uc774\ubbf8\uc9c0\uac00 \uc9c4\uc9dc\uc778\uc9c0 \uac00\uc9dc\uc778\uc9c0 \ubd84\ub958\ud569\ub2c8\ub2e4.<\/li>\n<\/ul>\n