/optimastic

Comparison of various stochastic optimization techniques

Primary LanguageC++MIT LicenseMIT

Optimastic!

Optimastic is a project to that aims to help the originator (Tarun) figure out how one should best structure stochastic optimizers. The goals of this project include:

  1. Figure out how to extend parallelization schemes (such as Hogwild) to stochastic optimizers with momentum
  2. Test out some of the new accelerated stochastic optimizers (such as Katyusha) that have theoretical promises
  3. Make it easy to perform experiments on new optimizers
  4. Make an interface that is thread-safe and easy to adapt to the massively parallel regime