Stochastic Optimization - Optimal selection of parameters for a model which contains uncertainty. Gradient-based search of solution space.
1. A set of locations in solution space are chosen.
2. The set of possible solutions to be evaluated in each location is partitioned into “baskets” or chunks for parallel evaluation.
3. The number of gradients per point in solution space is chosen.
4. Each “basket” can be evaluated in parallel.
5. Execution times are frequently measured in days.