Improving the Price of Anarchy via Predictions in Parallel-Link Networks

TL;DR

This paper explores how machine-learned predictions can improve coordination mechanisms in parallel-link congestion games, optimizing social costs with high accuracy and robustness even under prediction errors.

Contribution

It introduces a simple advice-based mechanism that is both consistent and robust, providing a full characterization and optimal robustness for monotone cost functions.

Findings

Mechanism achieves optimal robustness.

Advice improves social cost when predictions are accurate.

Mechanism maintains bounded performance under prediction errors.

Abstract

We study non-atomic congestion games on parallel-link networks with affine cost functions. We investigate the power of machine-learned predictions in the design of coordination mechanisms aimed at minimizing the impact of selfishness. Our main results demonstrate that enhancing coordination mechanisms with a simple advice on the input rate can optimize the social cost whenever the advice is accurate (consistency), while only incurring minimal losses even when the predictions are arbitrarily inaccurate (bounded robustness). Moreover, we provide a full characterization of the consistent mechanisms that holds for all monotone cost functions, and show that our suggested mechanism is optimal with respect to the robustness. We further explore the notion of smoothness within this context: we extend our mechanism to achieve error-tolerance, i.e. we provide an approximation guarantee that degrades smoothly as a function of the prediction error, up to a predetermined threshold, while achieving a bounded robustness.