AI 용어집
인공지능 완전 사전
Autoencoder
Unsupervised neural network composed of an encoder and a decoder that learns to efficiently compress and reconstruct input data.
Encoder
Part of the autoencoder responsible for compressing input data into a reduced-dimension representation in the latent space.
Decoder
Component of the autoencoder that reconstructs original data from their compressed representation in the latent space.
Latent space
Low-dimensional representation where compressed data by the encoder is stored, capturing essential features of input data.
Bottleneck
Minimal-dimension intermediate layer in an autoencoder that forces information compression and prevents direct copying.
Variational Autoencoder
Type of generative autoencoder that learns a probabilistic distribution in the latent space rather than a deterministic representation.
Denoising Autoencoder
Variant of autoencoder trained to reconstruct clean data from inputs corrupted by random noise.
Convolutional Autoencoder
Autoencoder using convolutional layers particularly effective for image and spatial data processing.
Sparse autoencoder
Autoencoder including a sparsity constraint to force the activation of only a few neurons in the hidden layer.
Contractive autoencoder
Autoencoder adding a penalty on the sensitivity of the representation to small variations in the input to improve robustness.
Reconstruction loss
Cost function measuring the difference between the original input data and their reconstruction by the autoencoder.
Deep autoencoder
Autoencoder architecture with multiple hidden layers allowing to learn hierarchies of complex features.
Memory autoencoder
Autoencoder integrating an external memory mechanism to store and retrieve learned representation prototypes.
Adversarial autoencoder
Combination of an autoencoder with a discriminator network to force the latent space to follow a specific distribution.
Reconstruction gap
Quantitative metric measuring the ability of an autoencoder to faithfully reconstruct the original input data.
Autoencoder overfitting
Phenomenon where the autoencoder memorizes the training data instead of learning generalizable representations.
Recursive Autoencoder
Autoencoder that processes hierarchical structures like syntactic trees by recursively applying encoding.
Hybrid Autoencoder
Architecture combining multiple types of autoencoders or integrating other deep learning models to improve performance.
Factor Disentanglement
Advanced objective of autoencoders aiming to separate independent factors of variation in the latent space.
Transformer Autoencoder
Autoencoder based on the transformer architecture using attention mechanisms to process sequential data.