Stochastic algorithms with geometric step decay converge linearly on sharp functions.
Damek DavisDmitriy DrusvyatskiyVasileios CharisopoulosPublished in: CoRR (2019)
Keyphrases
- orders of magnitude
- computationally efficient
- optimization problems
- theoretical analysis
- learning algorithm
- stochastic approximation
- computational efficiency
- monte carlo
- artificial neural networks
- computational complexity
- data structure
- evolutionary algorithm
- significant improvement
- computational cost
- worst case
- optimal solution
- objective function
- basis functions
- clustering algorithm
- recently developed
- genetic algorithm