Learning Payment-Free Resource Allocation Mechanisms

TL;DR

This work tackles the problem of allocating scarce, divisible resources among self-interested agents without monetary transfers, targeting welfare via Nash Social Welfare (NSW). It introduces ExS-Net, a modular neural architecture that simulates money-burning through a synthetic agent and optimizes an approximate incentive-compatibility objective by trading off exploitability against NSW, with an end-to-end training procedure and distribution-shift guarantees. The authors derive a generalization bound for the learning-based mechanism under finite samples and validate the approach through extensive experiments against state-of-the-art payment-free baselines, demonstrating substantial reductions in exploitability and competitive NSW and efficiency. The results suggest a practical, scalable path toward fair, incentive-aware resource allocation in settings where payments are infeasible, with robust performance under distributional shifts and untruthful data.

Abstract

We consider the design of mechanisms that allocate limited resources among self-interested agents using neural networks. Unlike the recent works that leverage machine learning for revenue maximization in auctions, we consider welfare maximization as the key objective in the payment-free setting. Without payment exchange, it is unclear how we can align agents' incentives to achieve the desired objectives of truthfulness and social welfare simultaneously, without resorting to approximations. Our work makes novel contributions by designing an approximate mechanism that desirably trade-off social welfare with truthfulness. Specifically, (i) we contribute a new end-to-end neural network architecture, ExS-Net, that accommodates the idea of "money-burning" for mechanism design without payments; (ii)~we provide a generalization bound that guarantees the mechanism performance when trained under finite samples; and (iii) we provide an experimental demonstration of the merits of the proposed mechanism.

Figures (8)

• Figure 1: The utility of agent 1 as it varies the reported preference ratio $v_{1,2}/v_{1,1}$ from 0.1 to 3. Blue line indicates the utility of agent 1 under truthful report, and red line indicates the maximum achievable utility under misreport. Exploitability of PF is shown as their gap. Agent $1$ increases its utility by under-reporting $v_{1,2}/v_{1,1}$.
• Figure 2: Performance trade-off frontier. The solid curve obtained by proposed mechanism ExS-Net over a range of exploitability tolerance (see Definition \ref{['def:epsilon_ic']}). The figure illustrates the superior exploitability/NSW trade-off achieved by ExS-Net over the mixture of PA and PF indicated by the dotted line (details in Section \ref{['sec:frontier']}).
• Figure 3: Exs-Net Pipeline
• Figure 4: Mechanism performance in 2x2 and 10x3 systems
• Figure 5: Mechanism performance in 40x3 and 60x2 systems

Theorems & Definitions (16)

• Definition 1: Nash Social Welfare
• Definition 2: Exploitability
• Definition 3
• Theorem 1: Generalization Bound
• Definition 4
• Definition 5
• Lemma 1: ?
• Lemma 2: Massart
• Lemma 3: ?
• Lemma 4: ?