Non-asymptotic estimates for TUSLA algorithm for non-convex learning with applications to neural networks with ReLU activation function.
Dong-Young LimAriel NeufeldSotirios SabanisYing ZhangPublished in: CoRR (2021)
Keyphrases
- learning algorithm
- neural network
- learning speed
- activation function
- feedforward neural networks
- genetic algorithm
- training process
- probabilistic model
- dynamic programming
- training phase
- feed forward neural networks
- input output
- recurrent neural networks
- rbf network
- prior knowledge
- error function
- pattern recognition