Glosarium AI
Kamus lengkap Kecerdasan Buatan
AnoGAN
Variant of GAN specifically designed for anomaly detection, using inverse mapping to reconstruct samples and calculate anomaly scores.
Generator
Neural network in a GAN that learns to generate synthetic samples by imitating the distribution of normal training data.
Discriminator
Neural network that distinguishes between real and generated samples, providing an error signal to improve the generator.
Adversarial loss function
Objective function that simultaneously optimizes the generator to fool the discriminator and the discriminator to correctly identify samples.
Adversarial training
Optimization process where the generator and discriminator mutually improve through continuous competition.
Normal distribution modeling
Ability of GAN to learn and represent the statistical distribution of normal data without anomalies.
Inverse mapping
Process to find the representation in latent space corresponding to a given sample, essential for calculating anomaly scores.
Conditional GAN
Extension of GANs where the generator and discriminator receive conditional information for more precise control of generation.
CycleGAN
GAN architecture with cyclic mappings used for anomaly detection through domain translation between normal and abnormal images.
Variational GAN
Combination of VAE and GAN providing better regularization of the latent space for more robust anomaly detection.
Detection threshold
Limit value of the anomaly score beyond which a sample is classified as abnormal, determined by statistical validation.
Bi-directional GAN
Architecture allowing both efficient generation and inference in the latent space, optimized for anomaly detection.