![[2601.01016] Improving Variational Autoencoder using Random Fourier Transformation: An Aviation Safety Anomaly Detection Case-Study](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2601.01016] Improving Variational Autoencoder using Random Fourier Transformation: An Aviation Safety Anomaly Detection Case-Study
Summary
This study explores enhancements to Variational Autoencoders (VAEs) using Random Fourier Transformation (RFT) for anomaly detection in aviation safety, demonstrating improved model performance.
Why It Matters
As aviation safety increasingly relies on advanced data analysis, this research provides insights into improving anomaly detection systems. By integrating RFT with VAEs, the study addresses critical challenges in identifying anomalies, which can enhance safety protocols and operational efficiency in the aviation industry.
Key Takeaways
- RFT improves training and inference in VAEs for anomaly detection.
- Models utilizing RFT can learn both low and high-frequency features simultaneously.
- The study introduces a trainable variant of RFT, enhancing model adaptability.
- Results indicate RFT models outperform conventional DNNs in anomaly detection.
- Further research is needed to assess the benefits of trainable RFT versus random variants.
Computer Science > Machine Learning arXiv:2601.01016 (cs) [Submitted on 3 Jan 2026 (v1), last revised 17 Feb 2026 (this version, v2)] Title:Improving Variational Autoencoder using Random Fourier Transformation: An Aviation Safety Anomaly Detection Case-Study Authors:Ata Akbari Asanjan, Milad Memarzadeh, Bryan Matthews, Nikunj Oza View a PDF of the paper titled Improving Variational Autoencoder using Random Fourier Transformation: An Aviation Safety Anomaly Detection Case-Study, by Ata Akbari Asanjan and 3 other authors View PDF HTML (experimental) Abstract:In this study, we focus on the training process and inference improvements of deep neural networks (DNNs), specifically Autoencoders (AEs) and Variational Autoencoders (VAEs), using Random Fourier Transformation (RFT). We further explore the role of RFT in model training behavior using Frequency Principle (F-Principle) analysis and show that models with RFT turn to learn low frequency and high frequency at the same time, whereas conventional DNNs start from low frequency and gradually learn (if successful) high-frequency features. We focus on reconstruction-based anomaly detection using autoencoder and variational autoencoder and investigate the RFT's role. We also introduced a trainable variant of RFT that uses the existing computation graph to train the expansion of RFT instead of it being random. We showcase our findings with two low-dimensional synthetic datasets for data representation, and an aviation safety datase...
