Glossario IA
Il dizionario completo dell'Intelligenza Artificiale
Reptile
Meta-learning optimization approach based on gradient descent that performs gradient descents on multiple tasks and updates the weights towards their average.
Bi-level Optimization
Two-level optimization structure used in meta-learning where the inner optimization adapts to tasks and the outer optimization learns to generalize.
Meta-Initialization
Weight initialization strategy for a neural network learned through meta-learning to facilitate rapid adaptation to new tasks.
Task-specific Adaptation
Process of fine-tuning parameters of a meta-learned model to adapt to the unique characteristics of a new task.
Task-agnostic Representation
Representations learned through meta-learning that capture general features useful for various tasks without being specific to any one of them.
Gradient-based Meta-Learning
Family of meta-learning algorithms that use gradients to directly optimize the rapid learning process.