[1I3-GS-5] Agents:

[1I3-GS-5-03] Efficient Preference Elicitation in Iterative Combinatorial Auctions with Many Participants

〇Ryota Maruo¹, Hisashi Kashima¹ (1. Kyoto University)

Keywords:Mechanism Design, Combinatorial Auction

A combinatorial auction is a type of auction where bidders can place bids or assign values to different subsets of goods (bundles).
One of the challenges in combinatorial auctions is the exponential increase in the number of possible subsets of goods as the number of items grows. Traditionally, it is assumed that bidders bid on every possible subset of goods, but this becomes impractical with a large number of items.
To address this issue, recent developments include the design of iterative combinatorial auctions using machine learning (ML). This method involves estimating each bidder's valuation function with ML models by asking them to value bundles, thereby seeking a more efficient allocation progressively. However, existing research typically involves training separate models for each bidder, which can be inefficient in cases where there are many bidders with similar valuation functions and a limited number of queries.
In our study, we have attempted to design a more efficient iterative combinatorial auction using multi-task learning, which allows for the sharing of information between models. This approach has proven to be more efficient in scenarios with many bidders or multiple bidders with similar valuation functions compared to existing studies.