Several emerging applications, such as "Analytics of Things" and "Integrative Analytics" call for a fusion of statistical learning (SL) and stochastic optimization (SO). The Learning Enabled Optimization paradigm fuses concepts from these disciplines in a manner which not only enriches both SL and SO, but also provides a framework which supports rapid model updates and optimization, together with a methodology for rapid model-validation, assessment, and selection. Moreover, in many big data/big decisions applications these steps are repetitive, and possible only through a continuous cycle involving data analysis, optimization, and validation. This paper sets forth the foundation for such a framework by introducing several novel concepts such as statistical optimality, hypothesis tests for model-fidelity, generalization error of stochastic optimization, and finally, a non-parametric methodology for model selection. These new concepts provide a formal framework for modeling, solving, validating, and reporting solutions for Learning Enabled Optimization (LEO). We illustrate the LEO framework by applying it to an inventory control model in which we use demand data available for ARIMA modeling in the statistical package "R". In addition, we also study a production-marketing coordination model based on combining a pedagogical production planning model with an advertising data set intended for sales prediction. Because the approach requires the solution of several stochastic programming instances, some using continuous random variables, we leverage stochastic decomposition (SD) for the fusion of regression and stochastic linear programming. In this sense, the novelty of this paper is its framework, rather than a specific new algorithm. Finally, we present an architecture of a software framework to bring about the fusion we envision.
Data Driven Decisions Lab at USC, Epstein Dept. of ISE, Univ. of Southern California, Los Angeles, CA 90089. March 14, 2017