TailBooster: A Dual-Layer Generative Framework for Extreme Value Augmentation with Operational Validity Enforcement
TailBooster targets rare flight disruptions by generating synthetic extremes, then filtering out records that do not fit learned operational constraints.
The paper says standard tabular generators miss distribution tails and can produce impossible flight records, such as short air times for long distances. TailBooster first isolates extreme cases with an interquartile-range layer, then trains dedicated Tabular VAE models on that tail-heavy signal. A second autoencoder layer removes synthetic samples outside the historical operational envelope. In US flight-record tests, the authors report lower extreme-event prediction error than conventional synthetic data, including 47-49% MAE reductions for extreme air time and 29-57% for extreme arrival delay. HF Daily Papers' note
The paper says standard tabular generators miss distribution tails and can produce impossible flight records, such as short air times for long distances. TailBooster first isolates extreme cases with an interquartile-range layer, then trains dedicated Tabular VAE models on that tail-heavy signal. A second autoencoder layer removes synthetic samples outside the historical operational envelope. In US flight-record tests, the authors report lower extreme-event prediction error than conventional synthetic data, including 47-49% MAE reductions for extreme air time and 29-57% for extreme arrival delay. HF Daily Papers' note
score 4