New Machine Learning Algorithm Enhances Combinatorial Auctions with Value and Demand Queries

A research paper introduces MLHCA, a novel machine learning-powered combinatorial auction that significantly improves efficiency by utilizing both value and demand queries. The study addresses the exponential growth challenge in bundle spaces for iterative combinatorial auctions (ICAs). While state-of-the-art ML-based algorithms currently rely on value queries for preference elicitation, practical ICAs employ demand queries. The new ML algorithm theoretically leverages full information from both query types, with experimental results demonstrating substantially better learning performance. This advancement represents a meaningful step in auction design, potentially impacting how preferences are elicited in complex bidding environments. The research was published on arXiv under identifier 2411.09355v3.