PROJECT ID: QNT-2025 // QUANTUM COMPUTING
Quantum Banknote Authentication
Our flagship quantum computing research: a hybrid quantum-classical machine learning model running on ODRA 5 — Poland's first superconducting quantum computer at Wrocław University of Science and Technology. The system authenticates banknotes by detecting subtle forgery patterns invisible to classical algorithms.
View RepositoryResearch Overview
The Challenge
Counterfeit detection requires identifying subtle, non-linear patterns in financial documents. Traditional machine learning struggles with the high-dimensional feature correlations present in wavelet-transformed banknote images. Our research explores whether quantum computing can provide a computational advantage for this security-critical task.
Our Approach
We implemented a Variational Quantum Classifier (VQC) that maps classical features into the quantum Hilbert space, enabling the detection of correlations that classical algorithms miss. The hybrid architecture combines quantum circuit execution on ODRA 5 with classical optimization via PyTorch's Adam optimizer.
Methodology
Data Pipeline
The UCI Banknote Authentication dataset provides 4 continuous features extracted from grayscale images via Wavelet Transform: variance, skewness, kurtosis, and entropy. Each feature is normalized and mapped to a qubit rotation angle, creating a quantum state that encodes the classical data in superposition.
Quantum Circuit
The 4-qubit Variational Ansatz consists of parameterized Ry gates for feature encoding followed by controlled rotation gates (CX) creating entanglement. Measurement in the Z-basis produces expectation values that are mapped to classification probabilities via a parity function.
Technical Stack
ODRA 5 System
Poland's first superconducting quantum computer, launched at PWr in 2025. Built for research in quantum informatics, telecommunications, and cybersecurity applications.
Qiskit ML
IBM's open-source quantum computing SDK with machine learning extensions. Provides circuit construction, simulation, and hardware execution interfaces.
Hybrid Training
PyTorch's Adam optimizer handles classical backpropagation while quantum gradients are computed via the Parameter Shift Rule on actual hardware.
Research Foundation
Based on Havlíček et al. (2019) quantum feature maps and Mitarai et al. (2018) variational classifier architecture.
Contributors
- Iwo Wojtakajtis
- Iwo Smura
- Rafał Balicki
- Karina Leśkiewicz
- Maria Płatek
- Michał Szczęsny