Tuesday, January 28, 2014 at 4:15pm
Frank H. T. Rhodes Hall, 253
ORIE Colloquium: Dragos Florin Ciocan (MIT) - Tools for Modern Revenue Management
We present potential solutions to several problems that arise in making revenue management practical for online advertising and related modern applications. Principally, such solutions must contend with highly volatile, hard to forecast demand processes, and massive scale making even basic optimization challenging. Our approaches to these problems are interesting in their own right in the areas of model predictive control, online packing and distributed optimization.
In the first part of the talk, we propose a model predictive control (MPC) approach to combat volatile demand. The scheme is conceptually simple, uses available demand data naturally, and can be shown to generate significant revenue advantages on real ad-network data. Under mild restrictions, we prove that our algorithm achieves uniform performance guarantees vis-a-vis a clairvoyant under arbitrary volatility while simultaneously being optimal if volatility is negligible.
The second part of the talk focuses on the problem of solving terabyte sized LPs given a distributed computational infrastructure; to this purpose, we design an LP algorithm that fits the 'Map-Reduce' paradigm. An implementation in a shared memory environment where we can benchmark against solvers such as CPLEX shows that the algorithm outperforms those solvers on the types of LPs arising in the online advertising context.