Glossario IA
Il dizionario completo dell'Intelligenza Artificiale
Dimensionality Compression
Process of reducing the number of variables while preserving relevant information, performed by the bottleneck layer of the autoencoder.
Reconstruction
Main objective of the autoencoder consisting of recreating the input data as faithfully as possible after compression and decompression.
Sparse Autoencoder
Autoencoder using a sparsity constraint on the hidden layer activations to activate only a few neurons simultaneously.
Reconstruction Loss Function
Metric measuring the difference between the original data and their reconstruction, typically mean squared error or binary cross-entropy.
Non-linear Encoding
Process of transforming data through non-linear activation functions, allowing to capture complex relationships in the data.
Latent Space Regularization
Technique aiming to impose constraints on the structure of the latent space to improve generalization and generative capabilities.
Stacked Autoencoder
Composition of several autoencoders trained sequentially to create increasingly abstract hierarchical representations.
Latent Space Resampling
Process consisting of navigating the latent space to explore or interpolate between different data representations.