Nearly-tight VC-dimension bounds for piecewise linear neural networks.
Nick HarveyChristopher LiawAbbas MehrabianPublished in: COLT (2017)
Keyphrases
- vc dimension
- piecewise linear
- upper bound
- lower bound
- neural network
- worst case
- vapnik chervonenkis
- covering numbers
- sample complexity
- statistical learning theory
- generalization bounds
- dynamic programming
- uniform convergence
- distribution free
- concept classes
- sample size
- empirical risk minimization
- inductive inference
- hyperplane
- concept class
- function classes
- pattern recognition
- optimal solution
- computational complexity
- objective function
- compression scheme
- data sets
- np hard