Kshitij Kulkarni

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2024-221

December 19, 2024

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2024/EECS-2024-221.pdf

This thesis studies problems that decision-makers face in environments that react or adapt to them. Chapter 2 studies incentive design problems where a decision-maker sets incentives over strategic agents who adapt over time and introduces a two-timescale algorithm for solving these problems. Applications include atomic networked aggregative games and non-atomic routing games. Chapter 3 considers the problem of inferring causal relationships between the state of a dynamical system and the occurrence of a rare event associated with the system and introduces an algorithm for solving this problem by aggregating data across time. Chapter 4 studies the formation of coalitions of charging stations in electric vehicle charging games and analyzes their performance relative to regimes where there are no coalitions and where there are coalitions of maximal size. Chapter 5 studies the redesign of congestion pricing schemes in settings where there are heterogeneities between the values-of-time of travelers during traffic routing. Chapter 6 looks at group decision-making problems in dynamic settings, where the preferences of agents might change, in contrast to the classical setting. In this setting, we prove a representation theorem that characterizes long-run policies.

Advisors: S. Shankar Sastry

---

BibTeX citation:

@phdthesis{Kulkarni:EECS-2024-221,
    Author= {Kulkarni, Kshitij},
    Title= {Algorithms for Learning and Incentive Design in Societal-Scale Systems},
    School= {EECS Department, University of California, Berkeley},
    Year= {2024},
    Month= {Dec},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2024/EECS-2024-221.html},
    Number= {UCB/EECS-2024-221},
    Abstract= {This thesis studies problems that decision-makers face in environments that react or adapt to them. Chapter 2 studies incentive design problems where a decision-maker sets incentives over strategic agents who adapt over time and introduces a two-timescale algorithm for solving these problems. Applications include atomic networked aggregative games and non-atomic routing games. Chapter 3 considers the problem of inferring causal relationships between the state of a dynamical system and the occurrence of a rare event associated with the system and introduces an algorithm for solving this problem by aggregating data across time.  Chapter 4 studies the formation of coalitions of charging stations in electric vehicle charging games and analyzes their performance relative to regimes where there are no coalitions and where there are coalitions of maximal size. Chapter 5 studies the redesign of congestion pricing schemes in settings where there are heterogeneities between the values-of-time of travelers during traffic routing. Chapter 6 looks at group decision-making problems in dynamic settings, where the preferences of agents might change, in contrast to the classical setting. In this setting, we prove a representation theorem that characterizes long-run policies.},
}

EndNote citation:

%0 Thesis
%A Kulkarni, Kshitij 
%T Algorithms for Learning and Incentive Design in Societal-Scale Systems
%I EECS Department, University of California, Berkeley
%D 2024
%8 December 19
%@ UCB/EECS-2024-221
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2024/EECS-2024-221.html
%F Kulkarni:EECS-2024-221