CAREER: Harnessing Prediction Engines and Non-Monetary Mechanisms for Real-Time Decision Making

Cornell University, Ithaca NY

Investigators

Abstract

CAREER: Harnessing Prediction Engines and Non-Monetary Mechanisms for Real-Time Decision Making Smart societal systems - on-demand transportation, smart supply-chain and logistics networks, cloud platforms, the smart grid, financial-processing networks, etc - are revolutionizing all aspects of our economic and social lives. All these systems face similar decision-making challenges due to uncertainty, complex state-spaces, combinatorial constraints, and strategic agent behavior. The aim of this Faculty Early Career Development Program (CAREER) project is to develop a unified framework for real-time decision-making for smart systems, built around the use of data-driven prediction oracles and non-monetary market mechanisms. Such an approach leads to policies that are simple, easy to interpret and implement in practice; understanding their performance however requires new theoretical and methodological ideas. The research will be complemented by collaborations with partners in on-demand transportation, cloud computing, online payment processing and the local food-bank, which will provide a portfolio of examples for pedagogical purposes. These collaborations tie in with outreach plans, which center on the development of a public library of societal systems simulators. These will provide research projects for undergraduate students, experiential problems for courses, and public demonstrations for attracting K-12 students to STEM fields. From a technical perspective, this research will develop rigorous frameworks for: (i) harnessing prediction oracles as inputs to real-time control policies, and (ii) designing non-monetary allocation policies based on emulating monetary mechanism. An exemplar of the paradigm is the idea of simulation-as-a-service (SaaS), wherein complex data-driven simulators will be used as inputs for control policies and mechanisms. Such an approach leverages historical and real-time data, incorporates the unique constraints of the underlying system, and results in simple and practical policies. The cost of this simplicity is that it is harder to prove rigorous guarantees. To overcome this, the research will couple advances in machine learning and mechanism design theory with the underlying philosophy of model-predictive control. In particular, new theoretical techniques, using ideas from stochastic coupling, convex optimization, martingale duality, and measure concentration, will be developed. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.