Distribution-free algorithms for predictive stochastic programming in the presence of streaming data

What is it about?

One of the major handicaps of Stochastic Programming is that such models require that the user specify a distribution associated with the data uncertainty. We recommend combining the input data process with a variety of online optimization algorithms such as stochastic quasi-gradient methods, as well as the stochastic decomposition (SD) algorithm. These results are the first of their kind in the stochastic optimization literature.

Featured Image

Why is it important?

Previous stochastic optimization methods require that the distribution be input to the algorithm, whereas, this new procedure learns the distribution of the data process as more data becomes available. Such automation can be very valuable for applications in which one encounters streaming data.