Michigan Researchers Win Both Best Paper Awards at AAMAS 2015
The two winning papers were selected from a field of 127 full paper submissions in the main technical track.
Michigan researchers won both best paper awards at the International Conference on Autonomous Agents & Multiagent Systems 2015, which took place May 4-8 in Istanbul, Turkey. The two winning papers were selected from a field of 127 full paper submissions in the main technical track.
Best Paper
Selected for the conference Best Paper Award was “The Dependence of Effective Planning Horizon on Model Accuracy,” which was authored by CSE graduate student Nan Jiang, CSE postdoctoral researcher Alex Kulesza, EECS Prof. Satinder Singh Baveja, and Psychology and Linguistics Prof. Richard L. Lewis.
The paper provides a precise explanation of why using a shorter planning horizon with a model estimated from data for Markov decision processes with long horizons can actually be better than a policy learned with the true horizon.
The researchers’ explanation for this phenomenon is based on principles of learning theory. They show formally that the planning horizon is a complexity control parameter for the class of policies to be learned. In particular, it has an intuitive, monotonic relationship with a simple counting measure of complexity, and that a similar relationship can be observed empirically with a more general and data-dependent Rademacher complexity measure. Each complexity measure gives rise to a bound on the planning loss predicting that a planning horizon shorter than the true horizon can reduce overfitting and improve test performance. They confirm these predictions empirically.
Pragnesh Jay Modi Best Student Paper
Selected for the Pragnesh Jay Modi Best Student Paper Award was “Welfare Effects of Market Making in Continuous Double Auctions,” which was authored by CSE graduate student Elaine Wah with her advisor, EECS Prof. Michael P. Wellman.
In the paper, the researchers investigate the effects of market making on market performance. They employed empirical simulation-based methods to evaluate heuristic strategies for market makers as well as background investors in a variety of complex trading environments.
Their findings indicate that the presence of the market maker strongly tends to increase not only total surplus across a variety of environments, but also background-trader surplus in thin markets with impatient investors, with urgency captured by a limited trading horizon. Comparison across environments reveals factors that influence the existence and magnitude of benefits provided by the market maker function.