Mean Field Equilibria of Dynamic Games: Theory and Computation

点击翻译按钮获取中文摘要

This project studies an approximation methodology known as mean field equilibrium (MFE), which can be used for management, design, and control of large dynamic economic systems such as financial markets, power markets, sponsored search auctions, online marketplaces,and resource sharing systems such as cloud computing platforms. In a mean field equilibrium, individuals postulate that fluctuations in the empirical distribution of otherplayers' states have "averaged out" due to large scale, and thusoptimize holding the state distribution of other players fixed. MFE simplifies system description, making optimization, design, andcontrol possible. Such a simplification in the equilibrium conceptdelivers the best of both worlds: not only is it more believable, butit provides a tool to understand the impact of parameter and designchanges on the performance of these complex dynamic systems. Thisproject develops methods to mathematically characterize regimes where MFEis a viable equilibrium concept, as well as to develop algorithms tocompute and learn MFE.If successful, the research results could be employed in practiceto leverage the huge advances in computation and availability of big data generated by complex economic systems for developingmore structured and principled prediction of the impact of designchanges. In support of this goal, the project includes efforts to make MFEtechniques accessible to engineering students through educationalmaterials and course curriculum.