DISTINGUISHED LECTURE SERIES: Autonomous Bidding Agents: Strategies and Lessons from the Trading Agent Competition

Abstract

In the design of autonomous trading agents that buy and sell goods in electronic markets, a variety of interesting computational questions arise. Perhaps the most fundamental is to determine how to bid on goods being auctioned off in separate markets when the agent's valuations for those goods are highly interdependent (i.e., complementary or substitutable). The Trading Agent Competition (TAC)--a challenging online trading game in the domain of travel shopping--was designed as a testbed in which to compare and contrast various approaches to this problem. Distilled from seven years of TAC experience, this talk recounts strategies and lessons (some theoretical and some experimental) to guide the design of autonomous bidding agents.