A PAC-Bayesian Approach to Spectrally-Normalized Margin Bounds for Neural Networks.
Behnam NeyshaburSrinadh BhojanapalliDavid McAllesterNathan SrebroPublished in: CoRR (2017)
Keyphrases
- neural network
- upper bound
- vc dimension
- pac bayesian
- mistake bound
- lower bound
- half spaces
- concept classes
- pattern recognition
- rademacher complexity
- sample complexity
- neural nets
- concept class
- pac learning
- worst case
- multilayer perceptron
- generalization error
- self organizing maps
- sample size
- similarity measure
- maximum margin
- upper and lower bounds
- distribution free
- support vector
- generalization bounds
- perceptron algorithm
- artificial neural networks
- fuzzy logic
- lower and upper bounds
- back propagation
- multi layer
- neural network model
- risk bounds
- training process
- covering numbers
- linear threshold
- training error
- statistical queries
- error bounds
- training set
- genetic algorithm