The large amount of aircraft trajectory data publicly available through open data sources like the OpenSky Network presents a wide range of possibilities for monitoring and post-operational analysis of air traffic performance. This contribution addresses the automatic identification of operational events associated with trajectories. This is a challenging task that can be tackled with both empirical, rule-based methods and statistical, data-driven approaches. In this paper, we first propose a taxonomy of significant events, including usual operations such as take-off, Instrument Landing System (ILS) landing and holding, as well as less usual operations like firefighting, in-flight refuelling and navigational calibration. Then, we introduce different rule-based and statistical methods for detecting a selection of these events. The goal is to compare candidate methods and to determine which of the approaches performs better in each situation.

No abstract

Deep Ensembles (DE) are a prominent approach achieving excellent performance on key metrics such as accuracy, calibration, uncertainty estimation, and out-ofdistribution detection. However, hardware limitations of real-world systems constrain to smaller ensembles and lower capacity networks, significantly deteriorating their performance and properties. We introduce Packed-Ensembles (PE), a strategy to design and train lightweight structured ensembles by carefully modulating the dimension of their encoding space. We leverage grouped convolutions to parallelize the ensemble into a single common backbone and forward pass to improve training and inference speeds. PE is designed to work under the memory budget of a single standard neural network. Through extensive studies we show that PE faithfully preserve the properties of DE, e.g., diversity, and match their performance in terms of accuracy, calibration, out-of-distribution detection and robustness to distribution shift.

No abstract