Prices, Bids, Values: One ML-Powered Combinatorial Auction to Rule Them All

Ermis Soumalias, Jakob Heiss, Jakob Weissteiner, Sven Seuken

TL;DR: Our ML-algorithm dramatically outperforms the SOTA for combinatorial auctions by combining demand queries and value queries.

Abstract:

We study the design of *iterative combinatorial auctions (ICAs)*.
The main challenge in this domain is that the bundle space grows exponentially in the number of items.
To address this, recent work has proposed machine learning (ML)-based preference elicitation algorithms that aim to elicit only the most critical information from bidders to maximize efficiency.
However, while the SOTA ML-based algorithms elicit bidders' preferences via *value queries*, ICAs that are used in practice elicit information via *demand queries*.
In this paper, we introduce a novel ML algorithm that provably makes use of the full information from both value and demand queries, and we show via experiments that combining both query types results in significantly better learning performance in practice. Building on these insights, we present MLHCA, a new ML-powered auction that uses value and demand queries. MLHCA significantly outperforms the previous SOTA, reducing efficiency loss by up to a factor 10, with up to 58% fewer queries.
Thus, MLHCA achieves large efficiency improvements while also reducing bidders' cognitive load, establishing a new benchmark for both practicability and efficiency. Our code is available at https://github.com/marketdesignresearch/MLHCA.

Lay Summary:

Imagine an auction where you want to buy a *combination* of items, not just one. These are called "combinatorial auctions". The challenge is that as more items are added, the number of possible combinations explodes, making it hard for buyers to express their preferences to the auctioneer. Therefore, the auctioneer needs to decide which items to assign to which buyers based on incomplete information, resulting in sub-optimal assignments.

Our research introduces a new approach using machine learning (ML) to make these complex auctions much more efficient. Traditionally, these auctions ask buyers about their value of specific combinations of items ("value queries") or what they would buy at different prices ("demand queries"). We've developed a novel ML algorithm, called MLHCA, that intelligently combines both types of queries.

By doing this, MLHCA significantly improves how well the auction learns buyers' true preferences, leading to better outcomes. In our experiments, MLHCA dramatically reduced the "efficiency loss" (how far the auction outcome is from the ideal) by up to a **factor of 10** and could match the previous state-of-the-art auction with up to **58% fewer questions** for buyers to answer. This means more effective auctions and less effort for participants.

For real-world applications of these auction formats like **spectrum auctions**, our simulations show that this improved efficiency could translate to approximately **100 million USD in additional value** per auction. MLHCA sets a new standard for how practical and efficient combinatorial auctions can be.

Application-Driven Machine Learning: This submission is on Application-Driven Machine Learning.

Link To Code: https://github.com/marketdesignresearch/MLHCA

Primary Area: Theory->Game Theory

Keywords: Combinatorial Auctions, Auction Design, Auctions, Market Design, Mechanism Design, Game Theory, Spectrum Auctions, Iterative Auctions, Preference Elicitation, Machine Learning, Neural Networks, Deep Learning, Bayesian Optimization, Active Learning, Computational Economics, Demand Queries, Value Queries

Submission Number: 8544