Foundations of decision making with computational and cognitive constraints: an ARO MURI (2019-24)
coPIs: Austin Benson, Joe Halpern, and Jon Kleinberg (Cornell). Josh Tenenbaum and Elchanan Mossel (MIT)
The goal of this new interdisciplinary research program is to create a rigorous foundation for individual and group decision-making under computational and cognitive constraints. The standard model of rational decision-making maintains that individuals use Bayes rule to incorporate new information into their beliefs. In addition to its normative appeal, this Bayesian paradigm serves as a highly useful benchmark by providing a well-grounded model. Despite these advantages, a growing body of evidence has scrutinized this framework on the basis of its unrealistic cognitive demand on individuals, especially when they make inferences in complex environments consisting of a large number of other decision-makers. To address these issues, researchers have adopted an alternative paradigm by assuming non-Bayesian behavior of agents. Over the past two decades, the fields of behavioral economics and behavioral decision theory have sought to explain observed deviations from the predictions of rational decision making of individuals and groups. Despite some success in such efforts, the modeling approaches that result are typically ad hoc and fail to articulate what deviations from Bayesian rationality actually lead to the observed non-Bayesian behavior in the agents. Consequently, we still lack a unified theory on human decision-making.
Our proposed effort is divided into three synergistic Thrusts, each composed of multiple interrelated tasks. In the first Thrust, we will develop a new foundation for individual decision making under computational and cognitive constraints and develop hierarchy of models that rationalize different cognitive biases. In the second Thrust, we will develop a unified framework for the study of rational and behavioral decision making in groups. Our third Thrust is devoted to data-driven modeling of biases as well as behavioral online experiments. Our team of 6 PIs from Cornell and MIT consists of distinguished experts in a wide range of areas, from network science to computational social science and algorithms, decision theory, game theory, cognitive science, and collective behavior.
Higher-order geometry and topology of complex networks. ARO Network Sciences Division
CoPIs: Austin Benson and Jon Kleinberg , Cornell
This research program is focused on development of novel theory, methods, and algorithms for higher order, non-dyadic interactions in complex networks and their functional implications. Complex networks underpin every area of current and future military and civilian infrastructure systems, and underpin integral parts of biological, physical, technological and socio-economic universe. Thus far, such networks have been mainly represented and analyzed as graphs with tools form graph theory used to analyze such systems. However, while graphs can capture pairwise interactions between nodes, fundamental interactions in networks often take place between multiple nodes. For example, in socio-economic networks, the joint coordinated activity of several agents (e.g. buyer, seller, broker); the formation and interactions of coalitions; and the existence of triadic closure are all prevalent.
The objective of this interdisciplinary proposal is to investigate how such interactions can be taken into account via the use of simplicial complexes (SCs), extensions of graphs that go beyond pairwise interactions and systematically account for interactions between groups of nodes (triplets, quadruplets, etc.). By introducing a novel notion of diffusion on SCs, we will extend key diffusion-based analysis techniques from graphs to the domain of SCs, thereby facilitating a more nuanced understanding of the studied systems. Specifically, we will study how to i) quantify the role edges in the information flow in a system; ii) predict group-based interactions; iii) use higher-order information for the detection of outliers and anomalies; and iv) detect mesoscale structures in higher-order data. We anticipate that this research will lead to a deepened understanding of the information-flow in networked systems, and could, e.g., lead to improved designs of communication structures in teams, or the detection of anomalies in communication patterns.
ONR Vannevar Bush Faculty Fellowship: “A New Paradigm for Analysis of Complex, Networked, Social and Engineering Systems”
As the principal investigator of this multi-year research program, Professor Jadbabaie proposes to develop a rigorous theory of information aggregation, strategic interaction, and systemic risk in complex networks. The proposed research will help us understand how individuals make decisions, how social phenomena spread in networks, and how the structure of underlying social networks can help us determine whether small, local shocks can have large systemic effects. The proposed plan, if successful, will also lead to the development of a new set of tools and methods for analysis of large scale networks that goes beyond graph-heretic approaches and uses tools from algebraic topology to discover structures and patterns in networks for link prediction, develop new sociometric tools for ranking edges, and higher order cliques in terms of their centralities.
DARPA Fundamental Limits of Learning (FunLoL) program: “Foundations of Scalable Statistical Learning”
This multi-year award supports the DARPA Fundamental Limits of Learning (FunLoL) program, which seeks to develop methodologies to evaluate the capabilities of learning system designs and guide practical implementations based on a formal understanding of the boundaries of their performance. Professor Jadbabaie leads this multidisciplinary approach, which entails developing a framework to capture the dependencies among the different entities generating large-dimensional data sets, employing graphical models to describe causal and time-dependent inference, developing a theory to understand limits to learning given sequential and incomplete information, and validating these tools on diverse, real-world applications.
ARO MURI: Evolution of Cultural Norms and Dynamics of Socio-Political Change
The events of the past two years in the Arab world have made it clear that questions related to political change, cultural dynamics, and societal transformations are not only of first-order importance for social science, but also central for a scientific approach to policy making and planning. While advances in traditional game theory, political economy, development economics and political science have enabled us to provide a posteriori analysis, understanding and predicting these events require a new set of theory, modeling, field experiments and algorithmic tools that are amenable to analysis of sociopolitical change. This will require analytical techniques and explicit modeling of conflicts of interest and possible cooperation between distinct parties. Since many of the central questions of political, cultural and societal change involve interactions among individuals and groups with different identities, this will also require advances in study of networks and collective phenomena. As a result, social science alone is far from satisfactory for addressing these issues.
What is needed is a multidisciplinary, analytical framework for analysis, prediction, and ultimately control of socio-political phenomena. To address this important void, we have brought together a world-class team of experts with a history of fruitful collaboration, who has been at the forefront of an interdisciplinary research agenda on this topic with expertise that spans mathematical systems theory, economics, political science, algorithmic and computational game theory, operations research, control theory, and network science. Our team members have been pioneers in development of a rigorous discipline of mathematical and computational social science that combines modeling, theory, empirical analysis, behavioral lab experiments, large datasets, and field experiments and surveys in diverse locations ranging from Afghanistan and India to the greater Middle East.
The Lagrange program seeks to develop new mathematical approaches to optimization problems in uncertain, dynamic, multiscale, and high-dimensional settings. By bridging methodologies developed for both discrete and continuous optimizations, Lagrange aims to enable solutions for complex, realistic problems that involve dynamic environments, rapidly changing requirements, and increasing or decreasing amounts of information.
In particular, Lagrange will address the fact that many applications of interest today are posed as non-convex optimization problems and thus remain intractable despite significant recent theoretical and algorithmic progress in convex optimization. Lagrange seeks methodologies beyond current convex relaxation methods to advance scalability of algorithms; data-driven approaches that explore proper sampling of data sets; and computationally tractable methods of approximating distributions.
Expected outcomes of the program include: 1) new mathematical frameworks and solution methods for large-scale optimization of complex systems, and 2) algorithms that could be implemented on computing platforms that would use parallelizability and scalability.