CVAE (条件变分自动编码器)

最新推荐文章于 2025-07-02 08:28:34 发布

「已注销」

最新推荐文章于 2025-07-02 08:28:34 发布

阅读量2.8k

点赞数

CC 4.0 BY-SA版权

文章标签： pytorch 深度学习 python

本文链接：https://blog.csdn.net/int_main_Roland/article/details/125112035

本文介绍了CVAE（条件变分自动编码器）的基本原理，包括模型结构、变分推断的SGVB框架以及如何优化ELBO。在实验中观察到，合适的超参数能加速收敛至优质模型，而过大或过小的参数可能导致不佳的模型性能。神经网络的学习过程呈现出先学数字轮廓后学细节的特点，且在训练过程中存在过拟合的风险。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

notations

$x$ image
$z$ latent
$y$ label (omitted to lighten notation)
$p (x ∣ z)$ decoder Encoder
$q (z ∣ x)$ encoder Decoder
$p^(h)\hat{p}(h)$ prior encoder (by variational inference) PriorEncoder

model structure

class CVAE(nn.Module):
    def __init__(self, config):
        super(CVAE, self).__init__()
        self.encoder = Encoder(...)
        self.decoder = Decoder(...)
        self.priorEncoder = PriorEncoder(...)
    def forward(self, x, y):
        x = x.reshape((-1, 784)) # MNIST
        mu, sigma = self.encoder(x, y)
        prior_mu, prior_sigma = self.priorEncoder(y)
        z = torch.randn_like(mu)
        z = z * sigma + mu
        reconstructed_x = self.decoder(z, y)
        reconstructed_x = reconstructed_x.reshape((-1, 28, 28))
        return reconstructed_x, mu, sigma, prior_mu, prior_sigma
    def infer(self, y):
        prior_mu, prior_sigma = self.priorEncoder(y)
        z = torch.randn_like(prior_mu)
        z = z * prior_sigma + prior_mu
        reconstructed_x = self.decoder(z, y)
        return reconstructed_x
#
class Loss(nn.Module):
    def __init__(self):
        super(Loss,self).__init__()
        self.loss_fn = nn.MSELoss(reduction='mean')
        self.kld_loss_weight = 1e-5
    def forward(self, x, reconstructed_x, mu, sigma, prior_mu, prior_sigma):
        mse_loss = self.loss_fn(x, reconstructed_x)
        kld_loss = torch.log(prior_sigma / sigma) + (sigma**2 + (mu - prior_mu)**2) / (2 * prior_sigma**2) - 0.5
        kld_loss