Prompt Description

A clean and professional deep learning architecture diagram of a Variational Autoencoder (VAE) is presented. On the left, an encoder network, composed of convolutional layers, progressively downsamples a 32×32 RGB image. The encoder outputs two vectors labeled “μ(x)” and “σ(x)”. In the center, a latent space block illustrates the reparameterization trick: z = μ + σ ⊙ ε, with ε sampled from a standard Gaussian distribution. On the right, a decoder network reconstructs the image using a robust architecture incorporating residual blocks, attention modules, and PixelShuffle upsampling layers, progressively increasing spatial resolution back to 32×32×3. Arrows indicate data flow from encoder to latent space to decoder. The design is minimalist and flat, with a white background, clear labels, and an academic style, suitable for a machine learning presentation.