Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
WardropNet: Traffic Flow Predictions via Equilibrium-Augmented Learning
Authors: Kai Jungel, Dario Paccagnan, Axel Parmentier, Maximilian Schiffer
ICLR 2025 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We present a comprehensive numerical study and show that Wardrop Net outperforms pure machine learning (ML) baselines on various realistic and stylized environments in both time-invariant and time-variant settings, yielding accuracy improvements of up to 75%. |
| Researcher Affiliation | Academia | Kai Jungel School of Management Technical University of Munich Munich, Germany EMAIL Dario Paccagnan Department of Computing Imperial College London London, United Kingdom EMAIL Axel Parmentier CERMICS Ecole des Ponts Marne-la-Vall ee, France EMAIL Maximilian Schiffer School of Management & Munich Data Science Institute Technical University of Munich Munich, Germany EMAIL |
| Pseudocode | Yes | Algorithm 1 Frank-Wolfe Algorithm Algorithm 2 Successive Averages Algorithm Algorithm 3 Bar-Gera s Algorithm Algorithm 4 Projection Methods |
| Open Source Code | Yes | The code for reproducing the results presented in this section is available at https://github. com/tum BAIS/ML-CO-pipeline-Traffic Prediction. |
| Open Datasets | Yes | The origin-destination pairs are extracted from a calibrated real-world MATSim scenario.1 1https://github.com/matsim-scenarios/matsim-berlin?tab=readme-ov-file |
| Dataset Splits | Yes | For all scenarios, we create 9 training instances, 5 validation instances, and 6 test instances. |
| Hardware Specification | Yes | We run the experiments on a computing cluster using 28-way Haswell-EP nodes with Infiniband FDR14 interconnect and 2 hardware threads per physical core. |
| Software Dependencies | No | The paper mentions 'MATSim (Horni et al., 2016)' but does not provide a specific version number. No other specific software versions are listed. |
| Experiment Setup | Yes | We run the training for a maximum of 20 hours, or 100 training epochs. |