Online Decision-Making using Prediction Oracles

The online allocation of scarce resources is one of the canonical problems in many fields of engineering. In this talk, I will re-examine basic online resource allocation, with the aim of building bridges between these problems and the ever-improving predictions provided my modern machine-learning methods. To this end, I will present a new Bayesian-learning inspired algorithm for online stochastic packing problems which achieves the first horizon and budget independent regret bounds for these settings. Surprisingly, the result stems from elementary underlying tools – LP sensitivity and basic concentration of measures.
Sid Banerjee is an assistant professor in the School of Operations Research and Information Engineering (ORIE) at Cornell, and a technical consultant at Lyft. His research is on stochastic modeling and control, and the design of algorithms and incentives for large-scale systems. He got his PhD in ECE from UT Austin, following which he was a postdoctoral researcher in the Social Algorithms Lab at Stanford. His work is supported by an NSF CAREER award, as well as grants from the NSF and ARL.