Enhancing Affine Maximizer Auctions with Correlation-Aware Payment

Researchers have introduced Correlation-Aware Affine Maximizer Auctions (CA-AMA), a novel framework that significantly improves revenue over classic VCG-based auctions when bidder valuations are correlated. The paper shows classic auctions can perform arbitrarily poorly, while CA-AMA reaches optimal revenue while preserving dominant-strategy incentive compatibility. A practical two-stage training algorithm finds approximate optimal CA-AMAs, with extensive experiments showing improved revenue and minimal violation of individual rationality constraints across various distributions.