Final Report Summary - ADVANCE-AGT (Algorithmic Game Theory: Foundations and New Directions)
During the three years of the project, the following main objectives were addressed:
• Study and perform research in state of the art AGT topics.
• Becoming more involved in the task of conferences organization.
• During the re-integration phase, in addition to continuing the work in the above two main objectives, a third goal has also been achieved: I designed and taught two courses on AGT at Technion in order to disseminate research ideas in AGT and to attract new graduate students to this research agenda.
Research in AGT topics: During the project, I studied and performed research in the following research topics: (1) optimal auction design, (2) the economic design of labor markets, (3) using auction design to improve the bitcoin protocol, (4) designing contests for revenue share, (5) deferred acceptance auctions. All research directions have yielded promising results. The research on labor markets has been accepted for publication. The research on contest design is currently under submission. The other research topics have yielded promising results, and I plan to further pursue them until completion.
The research on optimal auction design has started as a result of many interesting conversations with another visitor to UC Berkeley, Prof. Yang Cai from McGill University. In the last several years, many attempts have been made in order to advance the state of the art of this problem using new computational and algorithmic techniques. Prof. Cai has contributed to many of these recent results. Before my arrival to Berkeley, I did not have meaningful technical expertise in this extremely important sub-field. The conversations with Yang introduced me to the technical aspects of the subject, and helped me develop skills and insights to the problem. I have spent several months reading many relevant papers, developing my understanding and expertise. I have already obtained preliminary results which I believe will lay the foundation for additional interesting results in the near future, and which (I believe) are already sufficient to be published as a short note. More specifically, I study a setting with one buyer and two items. The buyer's value for each item is either L or H, and her value for a bundle of items is additive. I show a simple closed-form characterization of the optimal auction for this case: the optimal auction is deterministic, it is either a bundling auction or two separate single-item auctions. (The choice between the two formats depends on the underlying distribution of types.) When there are three possible values, the situation changes quite dramatically as shown by Hart and Reny (2011).Thus, moving from two values to three values seems like a very significant step, which I am currently pursuing.
The research on labor markets is an attempt to connect a computational model of auctions with classic economic theory. It studies a two-sided matching market with a set of heterogeneous firms and workers in an environment where jobs are secured by regulation. Without job security Kelso and Crawford have shown that stable outcomes and efficiency prevail when all workers are gross substitutes to each firm. We show that by introducing job security, stability and efficiency may still prevail, and even for a significantly broader class of production functions. This work is joint with Robert Kleinberg from Cornell, Hu Fu from University of British Columbia, and the game theorist Rann Smorodinsky from Technion. The integration of the algorithmic viewpoint with the economic/game-theoretic view point yielded, I believe, an interesting, novel model. I gave a presentation about this work at the 2014 Decentralization Conference that took place in June 2014 at Stanford, among several seminars in various universities.
The research on the bitcoin protocol is a collaboration with a Berkeley post-doc Or Sattath (who has now moved to the Hebrew University). Via conversations with him I became aware of an ongoing debate among Bitcoin’s developers regarding the best way to increase Bitcoin’s limited throughput (the maximal number of transactions that can be supported by the network every day). Currently, the throughput is being artificially limited by setting a maximal “block” size, so that the competing network servers will be able to charge a non-vanishing commission for publishing a transaction (commonly termed as preventing a “race to the bottom”). We have been examining other techniques, based on auction theory, to provide good revenue guarantees for the network servers while eliminating the artificial limitations on the throughput of the network. This research topic is currently ongoing. We have examined various auction techniques, obtained a good understanding of the exact nature of the problem, and we are currently evaluating using simulations the quality of our proposed solutions.
In the fourth research topic, I study capacity-constrained competition and analyze its resulting revenue shares, in a framework of contest design. The analysis in this project contrasts the near-symmetric case, where firms have similar supply sizes, and the extremely asymmetric case, where one large firm dominates the market. In particular, we show that while in the near-symmetric case simple contests provide near optimal equilibrium revenues for all sellers, in the asymmetric case a large firm can design more complicated contests that yield disproportionally low equilibrium revenues for its smaller opponents. This is joint work with Uri Feige from Weizmann Institute and Moshe Tennenholtz from Technion. This work has been finalized during the third year of the project and is now under journal submission.
The fifth research topic is a joint project with Vasilis Gkatzelis from Drexel university. This research cooperation started during the special semester on economics and computation at UC Berkeley. We study the performance guarantees of deferred acceptance auctions. This is a class of ascending auctions that currently receives attention due to its use in the US FCC auction. We show a new auction in this class that obtains an approximation ratio of O(log m) to the optimal social welfare for any arbitrary setting of single-minded bidders. This improves upon a pervious result of Gkatzelis and Roughgarden, and is the best possible that can be obtained by any deferred acceptance auction. We are currently trying to generalize this result for the case of multi-minded (single-value) bidders.
Conference organization: During the project I have taken part in several organizational roles of conferences. These took significant amount of my time, and are compatible with a stated goal of my fellowship, to accept administrative roles that are offered to more established researchers, on their way to become more senior in their scientific communities. The most significant role that I agreed to perform is to chair the CS program committee of the 7th Symposium on Algorithmic Game Theory (SAGT), and be part of its organizing committee. SAGT is an already established, well known conference, one of the major European venues for research papers on all subjects of algorithmic game theory. Chairing a PC is a very demanding role. The PC chair chooses a program committee, leads the discussions regarding various submitted papers, builds the program, edits the printed proceedings, etc. In 2014, we received 65 submissions. After a careful evaluation the committee that I chaired chose 25 papers to be presented at the conference. These papers were grouped into seven sessions: “Matching Theory”, “Game Dynamics”, “Games of Coordination”, “Networks/Social Choice”, “Markets and Auctions”, “Price of Anarchy”, “Computational Aspects of Games”, and “Mechanism Design and Auctions”. The conference was held at the University of Patras, Greece, in September 2014. The conference proceedings, that includes all papers presented at the conference, were published by Springer, in a book that I edited. Presenting authors of accepted papers were from many EU countries, as well as from around the world. After the conference has ended, I have agreed to co-edit, jointly with Dr. Martin Hoefer from MPI, a special issue of the journal Theory of Computing Systems for best papers from the conference as well as from SAGT’15. Work on this special issue is expected to continue throughout 2016.
In addition to this significant role, I also served on as a senior PC member at the 15th ACM conference on Economics and Computation (EC’14), as the Tutorials chair at EC’16, and on the PC of quite a few other conferences. Serving in these committees signals an important recognition by the AGT community as well as a major responsibility that cannot be avoided. It enabled me to gather experience in the various stages of the PC work, and in the other aspects of conference organization. While time-consuming, I view this entire activity as an extremely important contribution to the community, which should not be refused.
Design and Teach AGT courses at Technion: During the reintegration year (the third year of the project) I taught two courses at Technion that focus on subjects related to AGT. In the winter semester I have taught “Auction Theory” and in the spring semester I taught “Mechanism Design”. Both these courses are for advanced undergraduate and graduate students. The second course, “Mechanism Design” was taught for the first time this year at Technion, and I developed and designed it from scratch during the reintegration year. It lays down the basic theories and background that a student needs to know in order to conduct research in AGT. I believe that the fact that we now offer such courses at Technion will promote AGT research at Technion.
Attract new research students: Following the two courses that I gave I have made contact with two excellent students, Elisheva Shemesh and Jonathan Wagner, and I have started to discuss with them follow-up research to the research topics that I have performed during the project.
• Study and perform research in state of the art AGT topics.
• Becoming more involved in the task of conferences organization.
• During the re-integration phase, in addition to continuing the work in the above two main objectives, a third goal has also been achieved: I designed and taught two courses on AGT at Technion in order to disseminate research ideas in AGT and to attract new graduate students to this research agenda.
Research in AGT topics: During the project, I studied and performed research in the following research topics: (1) optimal auction design, (2) the economic design of labor markets, (3) using auction design to improve the bitcoin protocol, (4) designing contests for revenue share, (5) deferred acceptance auctions. All research directions have yielded promising results. The research on labor markets has been accepted for publication. The research on contest design is currently under submission. The other research topics have yielded promising results, and I plan to further pursue them until completion.
The research on optimal auction design has started as a result of many interesting conversations with another visitor to UC Berkeley, Prof. Yang Cai from McGill University. In the last several years, many attempts have been made in order to advance the state of the art of this problem using new computational and algorithmic techniques. Prof. Cai has contributed to many of these recent results. Before my arrival to Berkeley, I did not have meaningful technical expertise in this extremely important sub-field. The conversations with Yang introduced me to the technical aspects of the subject, and helped me develop skills and insights to the problem. I have spent several months reading many relevant papers, developing my understanding and expertise. I have already obtained preliminary results which I believe will lay the foundation for additional interesting results in the near future, and which (I believe) are already sufficient to be published as a short note. More specifically, I study a setting with one buyer and two items. The buyer's value for each item is either L or H, and her value for a bundle of items is additive. I show a simple closed-form characterization of the optimal auction for this case: the optimal auction is deterministic, it is either a bundling auction or two separate single-item auctions. (The choice between the two formats depends on the underlying distribution of types.) When there are three possible values, the situation changes quite dramatically as shown by Hart and Reny (2011).Thus, moving from two values to three values seems like a very significant step, which I am currently pursuing.
The research on labor markets is an attempt to connect a computational model of auctions with classic economic theory. It studies a two-sided matching market with a set of heterogeneous firms and workers in an environment where jobs are secured by regulation. Without job security Kelso and Crawford have shown that stable outcomes and efficiency prevail when all workers are gross substitutes to each firm. We show that by introducing job security, stability and efficiency may still prevail, and even for a significantly broader class of production functions. This work is joint with Robert Kleinberg from Cornell, Hu Fu from University of British Columbia, and the game theorist Rann Smorodinsky from Technion. The integration of the algorithmic viewpoint with the economic/game-theoretic view point yielded, I believe, an interesting, novel model. I gave a presentation about this work at the 2014 Decentralization Conference that took place in June 2014 at Stanford, among several seminars in various universities.
The research on the bitcoin protocol is a collaboration with a Berkeley post-doc Or Sattath (who has now moved to the Hebrew University). Via conversations with him I became aware of an ongoing debate among Bitcoin’s developers regarding the best way to increase Bitcoin’s limited throughput (the maximal number of transactions that can be supported by the network every day). Currently, the throughput is being artificially limited by setting a maximal “block” size, so that the competing network servers will be able to charge a non-vanishing commission for publishing a transaction (commonly termed as preventing a “race to the bottom”). We have been examining other techniques, based on auction theory, to provide good revenue guarantees for the network servers while eliminating the artificial limitations on the throughput of the network. This research topic is currently ongoing. We have examined various auction techniques, obtained a good understanding of the exact nature of the problem, and we are currently evaluating using simulations the quality of our proposed solutions.
In the fourth research topic, I study capacity-constrained competition and analyze its resulting revenue shares, in a framework of contest design. The analysis in this project contrasts the near-symmetric case, where firms have similar supply sizes, and the extremely asymmetric case, where one large firm dominates the market. In particular, we show that while in the near-symmetric case simple contests provide near optimal equilibrium revenues for all sellers, in the asymmetric case a large firm can design more complicated contests that yield disproportionally low equilibrium revenues for its smaller opponents. This is joint work with Uri Feige from Weizmann Institute and Moshe Tennenholtz from Technion. This work has been finalized during the third year of the project and is now under journal submission.
The fifth research topic is a joint project with Vasilis Gkatzelis from Drexel university. This research cooperation started during the special semester on economics and computation at UC Berkeley. We study the performance guarantees of deferred acceptance auctions. This is a class of ascending auctions that currently receives attention due to its use in the US FCC auction. We show a new auction in this class that obtains an approximation ratio of O(log m) to the optimal social welfare for any arbitrary setting of single-minded bidders. This improves upon a pervious result of Gkatzelis and Roughgarden, and is the best possible that can be obtained by any deferred acceptance auction. We are currently trying to generalize this result for the case of multi-minded (single-value) bidders.
Conference organization: During the project I have taken part in several organizational roles of conferences. These took significant amount of my time, and are compatible with a stated goal of my fellowship, to accept administrative roles that are offered to more established researchers, on their way to become more senior in their scientific communities. The most significant role that I agreed to perform is to chair the CS program committee of the 7th Symposium on Algorithmic Game Theory (SAGT), and be part of its organizing committee. SAGT is an already established, well known conference, one of the major European venues for research papers on all subjects of algorithmic game theory. Chairing a PC is a very demanding role. The PC chair chooses a program committee, leads the discussions regarding various submitted papers, builds the program, edits the printed proceedings, etc. In 2014, we received 65 submissions. After a careful evaluation the committee that I chaired chose 25 papers to be presented at the conference. These papers were grouped into seven sessions: “Matching Theory”, “Game Dynamics”, “Games of Coordination”, “Networks/Social Choice”, “Markets and Auctions”, “Price of Anarchy”, “Computational Aspects of Games”, and “Mechanism Design and Auctions”. The conference was held at the University of Patras, Greece, in September 2014. The conference proceedings, that includes all papers presented at the conference, were published by Springer, in a book that I edited. Presenting authors of accepted papers were from many EU countries, as well as from around the world. After the conference has ended, I have agreed to co-edit, jointly with Dr. Martin Hoefer from MPI, a special issue of the journal Theory of Computing Systems for best papers from the conference as well as from SAGT’15. Work on this special issue is expected to continue throughout 2016.
In addition to this significant role, I also served on as a senior PC member at the 15th ACM conference on Economics and Computation (EC’14), as the Tutorials chair at EC’16, and on the PC of quite a few other conferences. Serving in these committees signals an important recognition by the AGT community as well as a major responsibility that cannot be avoided. It enabled me to gather experience in the various stages of the PC work, and in the other aspects of conference organization. While time-consuming, I view this entire activity as an extremely important contribution to the community, which should not be refused.
Design and Teach AGT courses at Technion: During the reintegration year (the third year of the project) I taught two courses at Technion that focus on subjects related to AGT. In the winter semester I have taught “Auction Theory” and in the spring semester I taught “Mechanism Design”. Both these courses are for advanced undergraduate and graduate students. The second course, “Mechanism Design” was taught for the first time this year at Technion, and I developed and designed it from scratch during the reintegration year. It lays down the basic theories and background that a student needs to know in order to conduct research in AGT. I believe that the fact that we now offer such courses at Technion will promote AGT research at Technion.
Attract new research students: Following the two courses that I gave I have made contact with two excellent students, Elisheva Shemesh and Jonathan Wagner, and I have started to discuss with them follow-up research to the research topics that I have performed during the project.