Talks - Naveed Mahmud: Hybrid Quantum-Classical Machine Learning using Qiskit

Quantum-classical machine learning combines quantum computing capabilities with traditional ML approaches to potentially speed up specific machine learning tasks

Core quantum computing concepts essential for hybrid ML include:

• Qubits (quantum bits) that can exist in superposition states
• Entanglement between qubits creating strong correlations
• Quantum circuits with gates for manipulating qubit states

The framework combines:

• Classical pre-processing and dimension reduction (PCA, wavelet transform)
• Quantum feature mapping to encode classical data into quantum states
• Quantum classification using QSVC (Quantum Support Vector Classifier) or VQC (Variational Quantum Classifier)

Implementation details:

• Uses Qiskit (IBM’s open-source quantum computing SDK)
• Requires parameterized quantum circuits (ANSATZ)
• Employs quantum kernels for classification tasks
• Can work with both quantum simulators and real quantum hardware

Key findings from experiments:

• Quantum models achieved comparable accuracy (~70-72%) to classical methods
• VQC showed shortest training times among tested approaches
• Hard wavelet transform combined with quantum methods showed promise for large datasets
• Performance maintained even with reduced feature dimensions

Applications focused on sentiment analysis using:

• Twitter dataset (English)
• Bengali news portal dataset
• Text preprocessing and word embedding before quantum processing

Current limitations include:

• Most implementations run on simulators rather than real quantum hardware
• API transitions as quantum frameworks evolve
• Different quantum computer architectures may affect algorithm performance