AI-ordlista
Den kompletta ordlistan över AI
Variational Autoencoders (VAE)
Basic architecture of variational autoencoders using probabilistic distributions in the latent space for data generation.
Beta-VAE
Variant of the VAE introducing a beta parameter to control the trade-off between reconstruction and latent space regularization.
Conditional VAE (CVAE)
VAE conditioned by additional information allowing precise control of generation according to specified attributes.
Adversarial Autoencoders (AAE)
Architecture combining VAE and adversarial networks to impose arbitrary distributions on the latent space.
Vector Quantized VAE (VQ-VAE)
Autoencoder using vector quantization in the latent space for discrete representations and efficient reconstruction.
Hierarchical VAE
Multi-level architecture with hierarchical latent variables to capture complex structures at different scales.
InfoVAE
Variant based on the principle of maximum mutual information to improve generation quality and avoid posterior collapse.
Convolutional VAE
VAE specialized for image processing using convolutional layers to capture spatial structures.
Recurrent VAE
Architecture integrating recurrent networks (LSTM/GRU) for modeling and generating temporal sequential data.
VAE-GAN Hybrid
Combination of VAE and GAN using VAE for encoding and GAN to improve the visual quality of generations.
Denoising VAE
Variant trained to reconstruct clean data from noisy inputs to improve robustness and denoising.
Semi-Supervised VAE
Extension of the VAE for semi-supervised learning combining labeled and unlabeled data in a unified framework.