Machine-Learned Prediction Equilibrium for Dynamic Traffic Assignment

Joint work with Lukas Graf, Tobias Harks, and Kostas Kollias.

Abstract

We study a dynamic traffic assignment model, where agents base their instantaneous routing decisions on real-time delay predictions. We describe a general mathematical model which, in particular, includes the settings leading to IDE and DE. On the theoretical side we show existence of equilibrium solutions under some additional assumptions while on the practical side we implement a machine-learned predictor and compare it to other static predictors.

Publications

• Extended abstract: Proceedings of the AAAI Conference on Artificial Intelligence, 36(5), 5059-5067, DOI: 10.1609/aaai.v36i5.20438
• ArXiv Preprint: DOI: 10.48550/arXiv.2109.06713
• Journal Paper: “Prediction Equilibrium for Dynamic Network Flows”, Journal of Machine Learning Research, 24(310):1−33, 2023, https://jmlr.org/papers/v24/22-1446.html