Reproduce “A Non-autoregressive Multi-Horizon Flight Trajectory Prediction Framework with Gray Code Representation” on OpenResearch

About this paper

Dongyue Guo, Zheng Zhang, Zhen Yan, Jianwei Zhang, Yi Lin

This work introduces FlightBERT++, a non-autoregressive framework for multi-horizon flight trajectory prediction that utilizes Gray Code representation and a differential prediction paradigm. The approach substantially improves prediction accuracy and computational efficiency while eliminating outliers caused by high-bit misclassifications inherent in previous binary encoding methods.

Key insights

• Gray Code representation, with its single-bit changes between successive values, effectively mitigates high-bit misclassification issues seen in binary encoding, leading to more stable and accurate predictions.
• Predicting differential values of trajectory attributes enhances sequence stationarity and reduces the required bit length for encoding, further improving prediction robustness.
• A non-autoregressive encoder-decoder framework with a Horizon-aware Context Generator enables parallel multi-horizon forecasting in a single inference pass, significantly boosting computational efficiency and reducing error accumulation.

Summary by alphaXiv.