Rapid training of deep neural networks without skip connections or normalization layers using Deep Kernel Shaping.
James MartensAndy BallardGuillaume DesjardinsGrzegorz SwirszczValentin DalibardJascha Sohl-DicksteinSamuel S. SchoenholzPublished in: CoRR (2021)
Keyphrases
- neural network
- training process
- deep belief networks
- multi layer
- deep architectures
- pattern recognition
- multiple layers
- feedforward neural networks
- training algorithm
- feed forward neural networks
- kernel methods
- error back propagation
- deep learning
- unsupervised learning
- fuzzy logic
- feed forward
- back propagation
- training phase
- kernel function
- single layer
- artificial neural networks
- neural network training
- high dimensional
- support vector
- reinforcement learning