Details for Talk on: 05.12.2017

Speaker: Gianluca Brero

Title: Machine Learning-Based Combinatorial Auctions

Abstract: Combinatorial auctions (CAs) are auction formats where bidders are allowed to express bids on combinations of goods. One of the main challenges that arise in their design is that the value space grows exponentially in the number of goods, which can make it impossible for bidders to express their full value function. In this talk, I will present a new design paradigm for CAs using machine learning to help the auctioneer identify the most relevant information to elicit from the bidders. We design CAs based on this paradigm that are (1) computationally tractable, (2) provide bidders with good incentives to report their true values and (3) achieve high allocative efficiency.