Wide neural networks: From non-gaussian random fields at initialization to the NTK geometry of training.
Luís CarvalhoJoão Lopes CostaJosé MourãoGonçalo OliveiraPublished in: CoRR (2023)
Keyphrases
- random fields
- neural network
- non stationary
- autoregressive
- training process
- training algorithm
- conditional random fields
- markov random field
- maximum entropy
- feedforward neural networks
- parameter estimation
- random field models
- feed forward neural networks
- textured images
- pseudo likelihood
- multi layer perceptron
- artificial neural networks
- training set
- neural network model
- structured prediction
- three dimensional
- random field model
- gibbs sampler
- higher order
- gaussian distribution
- image segmentation
- nonparametric density estimation
- markov fields
- latent variables
- back propagation
- model selection
- probabilistic model
- k means
- computer vision