AI-ordlista
Den kompletta ordlistan över AI
GAN-based Augmentation
Technique using generative adversarial networks to synthesize realistic new training samples from a limited dataset. GANs learn the underlying data distribution to generate plausible and diverse examples.
Variational Autoencoders
Generative neural network architecture that learns a compressed latent representation of data before reconstructing or generating new samples. VAEs are particularly useful for creating controlled variations in feature space.
Feature Space Augmentation
Augmentation technique that operates directly in feature space rather than in pixel space or raw data. This approach allows for creating semantically consistent variations while preserving structural relationships between classes.
Adversarial Augmentation
Augmentation technique that uses adversarial perturbations to create robust samples that improve the model's resistance to attacks. This approach strengthens generalization by exposing the model to extreme but plausible variations.
AutoAugment
Machine learning method that automatically optimizes data augmentation policies to maximize model performance on a given validation set. This approach discovers adaptive, domain-specific augmentation strategies.
Cross-Domain Augmentation
Strategy that transfers and adapts augmentation techniques from one domain to another to enrich limited datasets. This approach leverages cross-domain knowledge to create relevant variations in data-scarce contexts.