Training of quantum circuits on a hybrid quantum computer

Area of Science:

• Quantum Computing
• Machine Learning
• Artificial Intelligence

Background:

• Generative modeling, a machine learning technique, is poised to leverage near-term quantum computers.
• Quantum-classical hybrid approaches are essential for harnessing quantum computational advantages.

Purpose of the Study:

• To implement and evaluate a data-driven quantum circuit training algorithm for generative modeling.
• To investigate the impact of different classical optimization strategies on quantum circuit training.

Main Methods:

• Utilized a quantum-classical hybrid machine for training.
• Employed parameterized quantum circuits on a trapped ion quantum computer.
• Applied Particle Swarm and Bayesian optimization for classical training.

Main Results:

• Successfully trained a high-dimensional universal quantum circuit on the Bars-and-Stripes dataset.
• Demonstrated that quantum hardware and classical optimization strategy critically influence convergence.
• Showcased the first successful training of a high-dimensional universal quantum circuit.

Conclusions:

• Hybrid quantum-classical learning schemes show promise for generative modeling.
• The choice of quantum hardware and classical optimizer significantly impacts training outcomes.
• Further research is needed to address challenges in hybrid quantum machine learning implementation.