Fundamental Limits of Data Analytics for Sequential SelectionWhen: Friday, June 21, 2013 at 4:00 pm
Where: DA 5th fl
Speaker: Lav R. Varshney
Organization: IBM Research: Thomas J. Watson Research Center
Fundamental Limits of Data Analytics for Sequential Selection
Emerging systems involving humans and machines use data analytics as subcomponents to help prioritize people’s energy. Whether considering information retrieval, human resource actions, urban planning, or child welfare protection, the goal is to use machine learning algorithms to rank tasks in order of value using a noisy set of predictive features. The ranked list is then used for sequential selection by human practitioners. This talk develops a stochastic model of analytics-enabled sequential selection, derives fundamental limits using the theory of concomitants of order statistics, and assesses limits in terms of system-wide performance metrics like screening effort and value of objects selected. Connections to sample complexity results for bipartite ranking are also developed.