A two-stage subspace trust region approach for deep neural network training.
Viacheslav DudarGiovanni ChierchiaÉmilie ChouzenouxJean-Christophe PesquetVladimir SemenovPublished in: EUSIPCO (2017)
Keyphrases
- trust region
- neural network training
- optimization method
- global optimum
- training algorithm
- optimization methods
- levenberg marquardt
- neural network
- column generation
- newton method
- feature space
- principal component analysis
- hessian matrix
- optimization algorithm
- line search
- log likelihood
- least squares
- evolutionary algorithm
- differential evolution
- back propagation
- simulated annealing
- learning rate
- decision trees
- data sets
- optimization procedure
- faster convergence
- global convergence
- metaheuristic
- genetic algorithm
- cost function