Figure 1
The procedure of quantum ensemble learning.
Figure 2
Brief explanation of a quantum support vector machine technique.
Figure 3
Brief explanation of a variational quantum classifier technique.
Figure 4
Brief explanation of a data-reuploading method.
Figure 5
Classification results from quantum classifiers and classical algorithms.
Contributions: J. Jung developed models for quantum classifiers, ensemble learning protocols, ran the experiments and analyzed the results.
Quantum Machine Learning Sep. 2022 – Feb. 2023
Quantum ensemble learning for classification with various quantum classifiers
⌜Developed quantum classifiers both in kernel-based method and parametrized quantum circuit-based method, quantum
support vector machine (QSVM), variational quantum classifier (VQC), and data re-uploading classifier.
Using above three quantum classifiers with superconducting qubits and trapped ion qubits, implemented ensemble
learning to solve binary classification problems using Torch, Pennylane, and Qiskit and compared with 8 different
classical algorithms.⌟
Supervised learning, a machine learning technique that utilizes input data paired with labels, has traditionally been explored using classical algorithms. However, with advancements in quantum computing, new quantum algorithms have emerged, including quantum support vector machines, variational quantum classifiers, data re-uploading methods, quantum neural networks, and quantum generative adversarial networks.
I developed quantum classifiers using both kernel-based methods and parameterized quantum circuits, including quantum support vector machines (QSVM), variational quantum classifiers (VQC), and data re-uploading classifiers. By implementing these quantum classifiers on superconducting qubits and trapped ion qubits based simulators, I applied ensemble learning to address binary classification problems using Torch, Pennylane, and Qiskit and compared their performance against eight different classical algorithms. For QSVM and VQC, I focused on optimizing the feature map and parameterized encoding circuits, while for the data re-uploading technique, I enhanced the efficiency of encoding methods using virtual Z gates and devised a compact encoding scheme.