Computational Stochastic Optimization and Learning

CASTLE Labs works to advance the development of practical, scalable models and algorithms for solving a wide range of applications that involve decisions under uncertainty. Core activities include:

  • Modeling – We have developed a powerful framework for modeling any sequential decision problem, spanning stochastic resource allocation problems, pure learning problems, and problems that involve both.  We have unified all of the strategies used in what we call the  “Jungle of stochastic optimization.”
  • Solutions – We have identified four classes of policies that span every solution procedure that has been proposed for any sequential decision/learning problem.  
  • Theory – We analyze the properties of algorithms, including convergence proofs, rate of convergence, bounds, and properties of estimators in machine learning.
  • Computation – We do extensive testing of algorithms on Tower, which is a 15 node compute system with over 400 threads, including three 64 thread machines with 3 TB of RAM.

Surrounding the core activities in methodology are laboratories focusing on major areas of application:

  • Optimal learning – This research addresses the challenges of collecting information, when information (observations, simulations, laboratory and field experiments) are expensive. See our new hOLMES Data Sciences website for learning in materials science.
  • Health sciences – Projects in health have included drug discovery, drug delivery, blood management, dosage decisions, personal health, and health policy.

I hope you find the material interesting, and perhaps useful. If you have any questions, please contact me.

Warren Powell