Title: Biddig with Upper and Lower Bounds in Machine Learning Powered Combinatorial Auctions
Abstract: In this talk, I present the ongoing research on machine learning powered iterative combinatorial auctions (CA). Preference elicitation is one major challenge in CAs due to the exponentially growing bundle space and potentially high costs in determining exact bundle values. The new auction addresses these issues in two ways: First, the auction mechanism uses a machine learning component to decide which bundle values a bidder must report at each round. Second, the bidder can report her values with some uncertainty (i.e. upper and lower bounds). Even though working with large value bounds, simulations show that the new auction creates higher efficiency compared to the well-known combinatorial clock auction.