High-Performance Quantum-Classical Hybrid Deep Generative Modeling Parameterized by Energy-based Models for Flight-Operations Anomaly Detection

The CIF project was a collaboration between Ames’s Quantum Artificial Intelligence Laboratory (QuAIL) and Data Sciences Group (DSG). Our team developed high-performance unsupervised deep machine-learning models for the detection of flight-operations anomalies. The models’ engineered-feature (latent) spaces are composed of discrete variables, which allows an integration with quantum computing because (part of) the latent-space variables can be populated by quantum-state measurements, which are discrete in nature.

Ames’s DSG had previously developed an unsupervised deep-learning model with continuous latent variables (variational autoencoder with Gaussian prior). The CIF project enabled the additional development of two quantum-capable variational autoencoders, with discrete latent space (Bernoulli and Boltzmann priors). The models exhibit state-of-the-art anomaly-detection performance and robustness when applied to aeronautics datasets. Future versions of our models will be deployed on in-time flight-operations data streams. They will also be used to assess the performance and resource requirements of quantum and other
physical computing devices.

The accurate and timely discovery of flight-operation anomalies is important because they can be precursors of potentially serious aviation incidents or accidents. To detect operationally significant anomalies and preempt future accidents, airlines and transportation agencies will have to increasingly rely on advanced machine-learning techniques applied to historical data or online data streams. Multifactorial and nonlinear anomalies defy traditional anomaly-detection methods, such as exceedance detection, and the volume and proportion of anomalies characterized by heterogeneous and high-order feature interactions are only expected to grow with increasing airspace complexity, characterized by increasing passenger volume, the integration of unmanned aircraft systems, and urban air mobility. Since labeled data (i.e. data classified as either nominal or anomalous) are costly to obtain and frequently not available and flight anomalies heterogeneous in nature, unsupervised learning approaches, such as the one portrayed in this paper, are often the preferred or only feasible option.

We developed high-performance unsupervised deep machine-learning models for the detection of anomalies in flight operations. The discovery of flight-operations anomalies is important because they can be precursors of potentially serious aviation incidents or accidents. The models’ engineered-feature spaces (latent spaces) are composed of discrete variables, which allows an integration with quantum computing because (part of) the latent-space variables can be populated by quantum-state measurements, which are discrete in nature.

This is early/mid-stage (TRL 3) — the most realistic path in is NASA SBIR/STTR, which funds small businesses and research institutions to develop technology aligned with NASA's needs (equity-free, phased funding). Check whether a current SBIR/STTR solicitation topic overlaps with this project's technology area, or contact the project directly (above) to ask.

None of these are guaranteed paths for this specific project — TechPort itself doesn't have an "apply" button. Reaching out to the contact(s) above with a specific question is usually the fastest way to find out what's actually open.