Competition between markets and the CAT Tournament: Guest editors' introduction to the special issue
Electronic Commerce Research and Applications
| Hi-index | 0.00 |
Resource allocation is a major issue in multiple areas of computer science. Auctions are commonly used in optimizing resource allocation in these areas, since well designed auctions achieve desirable economic outcomes including high allocative efficiency and fast response to supply and demand changes. This dissertation presents a grey-box approach to automated auction mechanism design using reinforcement learning and evolutionary computation methods. In contrast to the traditional approaches that try to design complete auction mechanisms manually, which is tedious and error-prone, the grey-box approach solves the problem through an automated search in a parameterized space of auction mechanisms. This space is defined by a novel, parameterized structure for auction mechanisms—a big white box—and a set of auction rules—each as a small black box—that can fit into the structure. The grey-box approach uses reinforcement learning to explore the composition of the structure, relates the performance of auction mechanisms to that of auction rules that form the mechanisms, and utilizes a Hall of Fame, a technique from evolutionary computation, to maintain viable auction mechanisms. The evaluation of auction mechanisms in the grey-box approach is conducted through a new strategic game, called CAT, which allows multiple marketplaces to run in parallel and compete to attract traders and make a profit. The CAT game helps to address the imbalance between prior work in this field that studied isolated auctions and the actual competitive situation that marketplaces face. Experiments were carried out to examine the effectiveness of the grey-box approach. A comparison against the genetic algorithm approach showed that the grey-box approach was able to produce mechanisms with significantly better overall performance. The best produced mechanisms from the grey-box experiments were able to outperform both the standard mechanisms which were used in evaluating sampled mechanisms during the grey-box search and carefully hand-coded mechanisms which won tournaments based on the CAT game. These best mechanisms also exhibited better performance than some existing mechanisms from the literature even when the evaluation did not take place in the context of CAT games.