Glossario IA
Il dizionario completo dell'Intelligenza Artificiale
Restricted Boltzmann Machines
Fundamental units of DBNs composed of bidirectional visible and hidden layers without intra-layer connections.
Greedy Pre-training
Sequential training technique where each RBM is trained independently before stacking.
Variational Inference
Approximation method for the posterior distribution in DBNs for efficient computation.
Deep Energy Models
Theoretical framework where DBNs define an energy function to model probability distributions.
Deep Autoencoders
Alternative architecture to DBNs using encoder-decoders for feature extraction.
Fine-tuning by Backpropagation
Supervised adjustment phase of pre-trained weights to optimize classification performance.
Hierarchical Feature Extraction
Multi-scale process where each layer captures increasingly abstract representations.
Deep Boltzmann Machines
Variants of DBNs with bidirectional connections between all adjacent hidden layers.
Generative Modeling
The ability of DBNs to generate new data by sampling from their learned distribution.
Optimization by Contrastive Divergence
Fast optimization algorithm for training RBMs in DBN architectures.
Hybrid Architectures
Combinations of DBNs with other models such as CNNs or RNNs for specific tasks.
Regularization and Dropout
Techniques to prevent overfitting in deep belief networks.