Teng and Wang achieve fair online resource allocation with anonymous pricing

A new paper from Yifeng Teng and Yifan Wang tackles a critical fairness issue in online resource allocation. Prior work by Ghuge, Singla, and Wang (GSW STOC'25) introduced a learnable dynamic pricing algorithm that maximized social welfare but required knowing the exact arrival order of heterogeneous agents in advance. This led to different agents facing different prices, raising equity concerns. The new research shows that anonymity does not have to come at the cost of optimality.

The authors demonstrate two key results. First, they prove that a polynomial number of samples from agents' value distributions is sufficient to learn the classic dual pricing algorithm—the same algorithm used in many practical settings. Second, they show that a polynomial number of pricing queries can learn a near-optimal anonymous pricing algorithm, where every agent sees item prices drawn from the same predetermined distribution. This removes the need for adversarial fairness debates and simplifies deployment, as the seller no longer requires knowledge of who arrives when.

The practical implications are significant. Online platforms that allocate limited resources (e.g., cloud compute, ad slots, energy) can now implement fair, welfare-maximizing pricing without complex tracking of user order or identity. The algorithms are sample-efficient and query-efficient, making them feasible for real-world use. This work bridges theoretical guarantees with practical fairness constraints, potentially influencing how companies like AWS, Google Cloud, and online marketplaces handle resource allocation.