AI-ordlista
Den kompletta ordlistan över AI
Latent Space Conditioning
Technique for injecting conditional variables into the latent space to control the characteristics generated by the CVAE.
Attribute Control
Mechanism enabling precise manipulation of generated data attributes by modifying the model's input conditions.
Conditional Prior
Modified prior distribution that depends on conditional variables, enabling guided sampling in the latent space.
Conditional Posterior
Posterior distribution that incorporates conditional information to infer relevant latent representations.
Class-Conditional VAE
Specialized variant of the CVAE using class labels as conditioning to generate samples from specific categories.
Multi-Modal Conditioning
Approach combining multiple types of conditions (text, image, audio) for granular control of generation.
Conditional Decoder
CVAE component receiving both the latent code and conditions to reconstruct or generate data.
Conditional Encoder
Neural network integrating conditions during encoding to produce a contextually informed latent representation.
Disentangled Representation
Latent representation where dimensions are independent and correspond to interpretable variational factors under conditioning.
Conditional Reconstruction Loss
Loss function measuring reconstruction fidelity while penalizing violations of specified conditions.
KL Divergence Conditioning
Adaptation of the Kullback-Leibler divergence accounting for conditional distributions in the training objective.
Conditional Sampling
Sampling technique in the latent space guided by conditions to generate targeted samples.
Feature Conditioning
Use of continuous or discrete features as conditions to refine control over generation.
Semi-Supervised CVAE
CVAE extension exploiting both labeled and unlabeled data simultaneously to improve conditional generation.
Hierarchical CVAE
Multi-level architecture incorporating conditions at different hierarchical scales for refined control.
Cross-Modal CVAE
Variant allowing conditioning of generation in one modality using information from another modality.
Adversarial CVAE
Hybridization combining conditional VAEs with adversarial networks to improve the quality of generated samples.
Conditional Interpolation
Interpolation technique in latent space preserving or gradually modifying specified conditions.