Scalable, Stochastic and Spatio-temporal Game Theory for Real-World
Human Adversarial Behavior
Prof. Milind Tambe Professor: Helen N. and Emmett H. Jones Professor in Engineering, Professor of Computer Science and Industrial and Systems Engineering Email: firstname.lastname@example.org Phone: (213) 740-6447 Office: PHE 410 (Powell Hall of Engineering)
For preventing terrorism at home or containing the activities of insurgent
networks overseas, it is critical to improve our understanding of adversarial behavior and develop
techniques to address it. Most current approaches assume players are isolated “rational
agents? when real-world human adversaries are distributed coalitions of socially, culturally and
cognitively-biased agents, acting behind a veil of uncertainty. We need fundamentally new gametheoretic
concepts and approaches based on more realistic models of player and play. The second
generation of operational systems for adversarial reasoning must address scalability, stochasticity
and spatiotemporal issues for which progress is inadequate or nonexistent.
Scalable Behavioral Game Theory
Human adversaries are rarely rational,
observant, computationally-capable entities devoted to a singular objective. We will develop
new scalable algorithms that leverage the bounded rationality and observations of humans, and address
human cognitive biases, cultural and social values as individuals or groups.
Stochastic Coalitional Game Theory
A real-world adversary is often a collection of distributed agents that
must communicate and coordinate to enact a joint action and whose the motivations may not be
perfectly aligned. This coordination in particular introduces new issues of stochasticity involving
coalition structures, information uncertainty, imperfect execution and robustness.
Spatiotemporal Game Theory
Practical adversarial domains involve acting in a geographical space over time
often in the midst of a population with divided loyalties. Incorporating spatiotemporal reasoning can substantially improve our ability to model and predict when and where consequential activities occur and how to address them.