Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Quantum Autoencoders for Anomaly Detection in Cybersecurity

Rohan Senthil, Swee Liang Wong

TL;DR

The paper investigates Quantum Autoencoders (QAEs) for anomaly detection in cybersecurity using the BETH dataset, comparing against Classical Autoencoders (CAEs) under data-limited conditions. It analyzes how quantum encoding (Amplitude, Angle, Dense-Angle, Efficient SU2), along with feature selection and ansatz choice, affects performance. The study finds that an 8-feature QAE with Dense-Angle encoding and a RealAmplitude ansatz delivers superior F1 scores (up to $0.87$) and separability ($0.97$) compared to CAEs, even with fewer training samples, and remains robust under simulated hardware noise. These results suggest QAEs can offer practical advantages for anomaly detection in data-limited cybersecurity contexts, given principled encoding and feature-engineering strategies.

Abstract

Anomaly detection in cybersecurity is a challenging task, where normal events far outnumber anomalous ones with new anomalies occurring frequently. Classical autoencoders have been used for anomaly detection, but struggles in data-limited settings which quantum counterparts can potentially overcome. In this work, we apply Quantum Autoencoders (QAEs) for anomaly detection in cybersecurity, specifically on the BPF-extended tracking honeypot (BETH) dataset. QAEs are evaluated across multiple encoding techniques, ansatz types, repetitions, and feature selection strategies. Our results demonstrate that an 8-feature QAE using Dense-Angle encoding with a RealAmplitude ansatz can outperform Classical Autoencoders (CAEs), even when trained on substantially fewer samples. The effects of quantum encoding and feature selection for developing quantum models are demonstrated and discussed. In a data-limited setting, the best performing QAE model has a F1 score of 0.87, better than that of CAE (0.77). These findings suggest that QAEs may offer practical advantages for anomaly detection in data-limited scenarios.

Quantum Autoencoders for Anomaly Detection in Cybersecurity

TL;DR

The paper investigates Quantum Autoencoders (QAEs) for anomaly detection in cybersecurity using the BETH dataset, comparing against Classical Autoencoders (CAEs) under data-limited conditions. It analyzes how quantum encoding (Amplitude, Angle, Dense-Angle, Efficient SU2), along with feature selection and ansatz choice, affects performance. The study finds that an 8-feature QAE with Dense-Angle encoding and a RealAmplitude ansatz delivers superior F1 scores (up to ) and separability () compared to CAEs, even with fewer training samples, and remains robust under simulated hardware noise. These results suggest QAEs can offer practical advantages for anomaly detection in data-limited cybersecurity contexts, given principled encoding and feature-engineering strategies.

Abstract

Anomaly detection in cybersecurity is a challenging task, where normal events far outnumber anomalous ones with new anomalies occurring frequently. Classical autoencoders have been used for anomaly detection, but struggles in data-limited settings which quantum counterparts can potentially overcome. In this work, we apply Quantum Autoencoders (QAEs) for anomaly detection in cybersecurity, specifically on the BPF-extended tracking honeypot (BETH) dataset. QAEs are evaluated across multiple encoding techniques, ansatz types, repetitions, and feature selection strategies. Our results demonstrate that an 8-feature QAE using Dense-Angle encoding with a RealAmplitude ansatz can outperform Classical Autoencoders (CAEs), even when trained on substantially fewer samples. The effects of quantum encoding and feature selection for developing quantum models are demonstrated and discussed. In a data-limited setting, the best performing QAE model has a F1 score of 0.87, better than that of CAE (0.77). These findings suggest that QAEs may offer practical advantages for anomaly detection in data-limited scenarios.
Paper Structure (22 sections, 6 equations, 6 figures, 2 tables)

This paper contains 22 sections, 6 equations, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Classical Autoencoders
  4. Quantum Autoencoders
  5. Quantum Encoding
  6. Dataset
  7. The BETH Dataset
  8. Data Preprocessing
  9. Experiments
  10. Quantum Autoencoder Architecture
  11. Encoding Techniques
  12. Amplitude Encoding
  13. Angle Encoding
  14. Dense-Angle Encoding
  15. Efficient SU2 Encoding
  16. ...and 7 more sections

Figures (6)

  • Figure 1: Diagram of a typical QAE
  • Figure 2: Comparison of two ansatz circuits: (a) RealAmplitude and (b) PauliTwoDesign. The features are encoded into $q0$ to $q3$ where $q0$ and $q1$ are the latent qubits and $q2$ and $q3$ are the trash qubits, $q4$ to $q5$ represent the reference qubits and $q6$ represents the auxiliary qubit.
  • Figure 3: Encoder Architecture for 24 input feature model
  • Figure 4: Performance comparison of QAEs and CAEs trained with 1,000, 5,000, 10,000, and 20,000 samples. CAE also includes the full training dataset. QAE was not trained with the full dataset due to resource constraints.
  • Figure 5: Score distributions of (a) the CAE simulated locally, (b) the QAE on a local noiseless simulation, and (c) the QAE on a local simulation with IBM Brisbane noise profile. While the CAE shows minimal separation, the QAE demonstrates a clear distinction between normal and abnormal data. The noise slightly shifts normal points, with no significant shift for anomalous data.
  • ...and 1 more figures