Conditional Variational Autoencoder (CVAE) for MNIST

Implementation of CVAE for generating and analyzing MNIST digits with controllable digit labels.

Features

• CVAE architecture with customizable latent dimensions
• Label conditioning via concatenation
• Multiple loss functions (BCE, MSE) comparison
• GPU support for M1/M2 MacBooks (MPS)
• Visualization tools

Results

• BCE loss outperformed MSE for digit generation
• Latent space visualization shows digit clusters
• Successful conditional generation of specified digits

Model Architecture

• Encoder: Input(784+10) → 512 → 256 → latent_dim
• Decoder: (latent_dim+10) → 256 → 512 → 784
• Label conditioning via one-hot encoding concatenation

Training Details

• Optimizer: Adam
• Learning rate: 1e-3
• Batch size: 128
• Epochs: 10
• Loss: BCE + KL divergence

Analysis Tools

• Loss curve plotting
• Latent space visualization
• Conditional sampling
• Multiple sample generation per digit

Dependencies

• PyTorch
• torchvision
• matplotlib
• numpy

Future Work

• Add quantitative metrics (FID score)
• Implement latent space interpolation
• Test different architectures