Jeff Hartline - Homepage

Jeff Hartline - Cornell University

Ultimate Frisbee (UPA, BULA, IAUA)

Primary research interest: Endogenous equilibrium selection in repeated games with an emphasis on the relationship between egality, stability, and long-term efficiency.

Ultimate	PhD Dissertation	Research Interests
Ultimate is non-contact sport played with 14 players, 8 cones, and a flying disc. The goal is to catch the disc in your team's endzone. You cannot run with the disc or it is a travel. You may not hold onto the disc for more than 10 seconds or it is a stall. You can throw the disc in any way you like, but if it is not caught by a teammate it is a turnover, and you must play defense to try to get the disc back. Often, you are required to lay-out in effort to help your team maintain possession or obtain possession of the disc. If you do not, you will and should be heckled mercilessly (unless you are new). If your lay-out is unsuccessful, it is probably still a nice bid. If your lay-out is successful, it might be sick, but mostly likely it was expected. Ultimate is played world-wide and year-round, by men and women, on grass and on sand, in the sun, wind, rain, and snow. It can even be played with a ball. Ultimate is self-refereed, governed by the spirit of the game, and sometimes aided by impartial observers who help make objective calls and resolve disputes. The motto of ultimate is make what should have happened happen, and continue play. If the involved players do not agree on what should have happened, do it over. This is a great system; no one cheats because everyone has the authority to call them on it, and everyone would rather play the game according to the rules .. because otherwise you're wasting your time. Players who make habitual unjustifiable calls are heckled relentlessly, and have no friends. My favorite throw is the blade. The blade is the best throw if you have it, and the worst throw if you don't.	My Cornell PhD dissertation is titled Incremental Optimization. My advisor was Dexter Kozen. A lot of my research was done jointly with Alexa Sharp. My dissertation defined a general model for incrementalizing a broad class of optimization problems and developed techniques for adapting approximation algorithms for single-level problems to k-level incremental versions of the same problem. An incremental problem is a problem for which the constraints develop in some monotonic fashion over some number of levels. An incremental solution is a sequence of solutions that build on each other incrementally. The goal of an incremental algorithm is to construct a solution that is a good solution at all levels. For typical incremental problems, this task is challenging because there is tension between local optimality and global optimality. Incremental problems are similar to stochastic problems and online problems. Stochastic problems are probabalistic: developing constraints obey some probability function and we typically optimize the expected value of the final solution. Online problems are adversarial: developing constraints are completely unknown, and we typically optimize the worst-case relative performance of the final solution. In my model of incremental problems, I assume that the changing constraints are deterministic and known from the outset. The goal is to optimize some function of a solution's performance at all levels. The price of knowledge is the difference between the performance of the best online algorithm and the best incremental algorithm for the same problem -- it is how much an optimizer would pay for information about the developing constraints. Please see my dissertation for more details.	In conventional game theory, players are naïvely rational responders to their environment. Many Nash equilibria are plausible, and a subset of these equilibria might be stable outcomes of players' learning strategies. Folk Theorem establishes that any feasible and rational outcome in a one-shot game G is a Nash equilibrium outcome for the repeated game G*. The realization of these equilbria involve a collectively followed plan and collectively enforced punishments for deviators. In practice, there is no exogenously provided plan, and agent strategies are an endogenous influence on stable outcomes. What happens if agents are not naïvely rational, but are instead willing to sacrifice local optimality for directing the repeated game towards a more desirable equilibrium? What game properties allow a consortium of cooperators to induce cooperation among myopic or manipulation- resistant players? To what extent must these consortia just accept a certain degree of free- loading? What roles do altruism, retaliation, and bluffing play in reputation development? How do reputations influence player strategy? How do player asymmetries, such as a power imbalance, influence the impact of manipulative strategies? Can exploited majorities defend themselves from powerful minorities? Is irrationality a reasonable strategy in exploitative circumstances? Is there a correlation between fairness and stability? As with my PhD dissertation, I am concerned with tensions between short-term and long-term optimality. When can we make the behavior of myopic agents compatible with long-term utility and global egality. I'm interested in the tradeoffs between competition and cooperation, efficiency and stability.

Ultimate

PhD Dissertation

Research Interests

Ultimate is non-contact sport played with 14 players, 8 cones, and a flying disc. The goal is to catch the disc in your team's endzone. You cannot run with the disc or it is a travel. You may not hold onto the disc for more than 10 seconds or it is a stall. You can throw the disc in any way you like, but if it is not caught by a teammate it is a turnover, and you must play defense to try to get the disc back. Often, you are required to lay-out in effort to help your team maintain possession or obtain possession of the disc. If you do not, you will and should be heckled mercilessly (unless you are new). If your lay-out is unsuccessful, it is probably still a nice bid. If your lay-out is successful, it might be sick, but mostly likely it was expected. Ultimate is played world-wide and year-round, by men and women, on grass and on sand, in the sun, wind, rain, and snow. It can even be played with a ball.

Ultimate is self-refereed, governed by the spirit of the game, and sometimes aided by impartial observers who help make objective calls and resolve disputes. The motto of ultimate is make what should have happened happen, and continue play. If the involved players do not agree on what should have happened, do it over. This is a great system; no one cheats because everyone has the authority to call them on it, and everyone would rather play the game according to the rules .. because otherwise you're wasting your time. Players who make habitual unjustifiable calls are heckled relentlessly, and have no friends.

My favorite throw is the blade. The blade is the best throw if you have it, and the worst throw if you don't.

My Cornell PhD dissertation is titled Incremental Optimization. My advisor was Dexter Kozen. A lot of my research was done jointly with Alexa Sharp.

My dissertation defined a general model for incrementalizing a broad class of optimization problems and developed techniques for adapting approximation algorithms for single-level problems to k-level incremental versions of the same problem. An incremental problem is a problem for which the constraints develop in some monotonic fashion over some number of levels. An incremental solution is a sequence of solutions that build on each other incrementally. The goal of an incremental algorithm is to construct a solution that is a good solution at all levels. For typical incremental problems, this task is challenging because there is tension between local optimality and global optimality.

Incremental problems are similar to stochastic problems and online problems. Stochastic problems are probabalistic: developing constraints obey some probability function and we typically optimize the expected value of the final solution. Online problems are adversarial: developing constraints are completely unknown, and we typically optimize the worst-case relative performance of the final solution. In my model of incremental problems, I assume that the changing constraints are deterministic and known from the outset. The goal is to optimize some function of a solution's performance at all levels. The price of knowledge is the difference between the performance of the best online algorithm and the best incremental algorithm for the same problem -- it is how much an optimizer would pay for information about the developing constraints.

Please see my dissertation for more details.

In conventional game theory, players are naïvely rational responders to their environment. Many Nash equilibria are plausible, and a subset of these equilibria might be stable outcomes of players' learning strategies. Folk Theorem establishes that any feasible and rational outcome in a one-shot game G is a Nash equilibrium outcome for the repeated game G*. The realization of these equilbria involve a collectively followed plan and collectively enforced punishments for deviators. In practice, there is no exogenously provided plan, and agent strategies are an endogenous influence on stable outcomes. What happens if agents are not naïvely rational, but are instead willing to sacrifice local optimality for directing the repeated game towards a more desirable equilibrium?

What game properties allow a consortium of cooperators to induce cooperation among myopic or manipulation- resistant players? To what extent must these consortia just accept a certain degree of free- loading? What roles do altruism, retaliation, and bluffing play in reputation development? How do reputations influence player strategy? How do player asymmetries, such as a power imbalance, influence the impact of manipulative strategies? Can exploited majorities defend themselves from powerful minorities? Is irrationality a reasonable strategy in exploitative circumstances? Is there a correlation between fairness and stability?

As with my PhD dissertation, I am concerned with tensions between short-term and long-term optimality. When can we make the behavior of myopic agents compatible with long-term utility and global egality. I'm interested in the tradeoffs between competition and cooperation, efficiency and stability.

To contact, please email: jhartlin AT cs DOT cornell DOT edu

Jeff Hartline - Cornell University

Publications

Cornell Dissertation

Curriculum Vitae (research)

Curriculum Vitae (corporate)

Ultimate Frisbee (UPA, BULA, IAUA)

Primary research interest: Endogenous equilibrium selection in repeated games with an emphasis on the relationship between egality, stability, and long-term efficiency.