AI 용어집
인공지능 완전 사전
Discrete Latent Variables
Latent variables taking discrete values from the codebook, enabling more interpretable and efficient modeling of data structures.
Encoder
Neural network transforming input data into continuous representations that are subsequently quantized into discrete indices from the codebook.
Decoder
Neural network reconstructing original data from quantized latent representations extracted from the codebook.
PixelCNN
Convolutional neural network often used to model the prior distribution of discrete latent codes in VQ-VAEs.
Latent Space Discretization
Process of transforming a continuous latent space into a discrete set of representations using vector quantization.
Hierarchical VQ-VAE
Multi-scale VQ-VAE architecture using multiple levels of quantization to capture hierarchical structures in data.
Vector Quantization Layer
Specific neural network layer implementing the vector quantization operation and selection of nearest codebook embeddings.