AI Glossary
The complete dictionary of Artificial Intelligence
Architecture Search Space
Set of all possible neural network architectures that can be generated by the NAS process, defined by constraints on layer types, connections, and hyperparameters.
Reinforcement Learning NAS
NAS approach using an RL agent that learns to generate network architectures by maximizing a reward based on the validated performance of the model on the target task.
HyperNetwork
Neural network that generates the weights of other networks, used in NAS to parameterize the architecture space and enable continuous optimization of both structure and weights simultaneously.
Early Stopping in NAS
Technique for prematurely stopping the training of candidate architectures based on predictors of final performance, significantly reducing the computational cost of evaluation.
Architecture Parameterization
Mathematical representation of the architecture space in NAS, using continuous, discrete, or hybrid variables to encode structural decisions and enable their algorithmic optimization.