A PAC-Bayesian Approach to Spectrally-Normalized Margin Bounds for Neural Networks.
Behnam NeyshaburSrinadh BhojanapalliNathan SrebroPublished in: ICLR (Poster) (2018)
Keyphrases
- neural network
- upper bound
- vc dimension
- pac bayesian
- mistake bound
- rademacher complexity
- pattern recognition
- lower bound
- worst case
- back propagation
- artificial neural networks
- fuzzy logic
- sample complexity
- multilayer perceptron
- error bounds
- risk bounds
- support vector
- average case
- neural nets
- neural network model
- self organizing maps
- concept classes
- pac learning
- training error
- upper and lower bounds
- distribution free
- worst case bounds
- generalization error
- feed forward
- training set
- learning algorithm
- genetic algorithm
- multi layer
- fault diagnosis
- sample size
- model selection
- support vector machine
- decision trees