SGD learning on neural networks: leap complexity and saddle-to-saddle dynamics.
Emmanuel AbbeEnric Boix AdseràTheodor MisiakiewiczPublished in: COLT (2023)
Keyphrases
- neural network
- learning process
- learning algorithm
- active learning
- artificial neural networks
- neural nets
- learning systems
- recurrent networks
- computational complexity
- inductive inference
- error back propagation
- stochastic gradient descent
- critical points
- incremental learning
- knowledge acquisition
- fuzzy logic
- pattern recognition
- multiscale