Metric-Aware Hybrid Forecasting for the CTF4Science Lorenz Challenge
2026-06-02 • Machine Learning
Machine LearningArtificial Intelligence
AI summaryⓘ
The authors participated in a challenge that required predicting and analyzing complex patterns from the Lorenz system, a well-known chaotic model. They found that no single method worked best for all parts of the task, so they combined different approaches tailored to each specific goal, such as reconstructing full paths, making short-term forecasts, and matching long-term data patterns. Their mix of techniques performed well on the leaderboard, and they shared a simpler version of their method for easier understanding and replication. Overall, the authors showed that combining specialized tools can be more effective than relying on one model alone.
Lorenz systemforecastingtrajectory reconstructionODE fittingdenoisersdistribution matchingtime series predictionsynthetic databenchmark challenge
Authors
Cen Lu
Abstract
We describe our approach to the CTF4Science Lorenz challenge, a benchmark that mixes short-horizon forecasting, long-time distribution matching, and trajectory reconstruction across nine task pairs. The key discovery is that no single model family dominated all metrics. Instead, we built a metric-aware hybrid system that assigned a different predictor to each metric family: (1) synthetic-pretrained denoisers for full-trajectory reconstruction, (2) Lorenz ODE fitting and trajectory shooting for the first 20 forecast steps, and (3) histogram-tail substitution using synthetic Lorenz libraries for long-time evaluation. A representative mature submission from this system family scored 83.83551 on the public leaderboard, and a small follow-up stack of the same ideas reached 83.85529. We focus on the cleaner intermediate system because it captures the full method while remaining simple enough to reproduce and analyze, while the final submission can be understood as a conservative extension of the same backbone.