NeTS: Small: PTERA: Prospect Theory Enhanced Resource Allocation

University Of California-Berkeley, Berkeley CA

Investigators

Abstract

Numerous empirical studies have shown that humans evaluate utilities in ways that are not consistent with utility theory. Cumulative prospect theory, pioneered by the Nobel prize-winning economist Daniel Kahneman, and his collaborator, Amos Tversky, provides a framework, including utility theory as a special case, which is able to capture a wide range of actually observed human behaviors when faced with choices. This project develops the implications of prospect theoretic models for how humans evaluate choice in the context of some of the core problems of communication networking. The project studies how communication networking protocols may need to be modified when the evaluation of choice is done by prospect-theoretic end users. It is expected that better network protocols that improve the end-user experience could be a result of this research. In cumulative prospect theory, gains and losses are measured relative to a reference. In order to make the theory consistent with empirically observed behavior the function used to evaluate gains and losses is concave for gains and convex for losses, with a larger slope for losses than for gains near the reference. A prospect is a list of possible gains or losses, each with a certain probability. Users are presented with prospects and modeled as evaluating them via the rules of cumulative prospect theory, which also involve weighting functions that overweight small probabilities relative to large probabilities. This project focuses on two specific, widely applicable, and highly important problem classes in communication networking. One of these is the question of how to incentivize users to follow best practices to secure their individual devices that are connected to the network. The project studies how to design the appropriate prospects to present to the users to achieve the goal of a better secured and therefore safer and more resilient network. The second of these is the problem of resource allocation in the network for elastic users. The project studies how to design TCP-like protocols that are backward compatible with the existing ones, but that will improve the perceived utility for prospect-theoretic users while still working with the same network resources. By trading off user experience of elastic users with the network resources used, this project also develops ways to improve the performance for inelastic users. Broader Impacts: Techniques for decision making and game theoretic techniques provide the underpinnings for a broad range of disciplines that study how humans interact with technological systems and with each other. Apart from communication networking, examples of such disciplines include economics and finance. The outcomes of this project have broader impact in these disciplines, because novel methodologies for cumulative prospect theory research that can be used for analysis and design in similar scenarios that are relevant to these individual disciplines are developed in this research. This project also has a significant educational component, including training graduate students and providing a platform for undergraduates to have a research experience. The research developed in this project will be incorporated into the curriculum of the course on estimation and control of stochastic systems and is expected to impact the course on network economics, both taught at Berkeley. The research provides opportunities for undergraduate projects, in particular through the Berkeley EECS departmental initiatives that foster diversity and inclusion by inviting a diverse group of undergraduates during the summers to participate in the research programs of the faculty members.