New matching algorithm guarantees perfect accuracy with prediction, 3x worst-case bound

Researchers Jabari Hastings and Marena Richter have solved a major open question in algorithm design by introducing a learning-augmented matching mechanism that works with only ordinal preferences. The classic problem involves matching n agents to n items on an unknown line metric, where each agent ranks items by distance. Without any additional information, the best possible distortion (ratio of algorithm cost to optimal cost) is 3. The new paper shows that by accepting a single predicted matching as a hint, their mechanism achieves 1-consistency (matching the optimal cost when the prediction is correct) while preserving the optimal 3-robustness guarantee even if the prediction is arbitrarily wrong.

This result directly addresses the open problem posed by Filos-Ratsikas et al. at IJCAI 2025. The key insight is that the mechanism uses the predicted matching to guide its decisions but falls back to a robust strategy when the prediction conflicts with the ordinal preferences. The algorithm requires no actual distance measurements—only ranked lists from each agent. This makes it practical for real-world scenarios where distances are unknown but preferences are easy to collect.

The work has immediate implications for market design, ride-sharing platforms, and task assignment systems where predictions about optimal matches can be derived from historical data. By proving that a single prediction can yield both perfect accuracy and theoretical worst-case guarantees, Hastings and Richter open the door for more reliable learning-augmented algorithms in online settings. The full paper is available on arXiv (2607.09038).