Automatic Generation of Optimized Quantum Circuits for Real-world Machine Learning Applications

Designing quantum circuits requires defining the circuit structure, the gates within that structure, and any possible gate parameterization. Currently, there are no generic methods for designing quantum circuits, and the process is often manual or based on predefined circuits that, in many cases, are not well suited to specific data sets. This seminar presents a methodology based on multiobjective genetic algorithms to automatically generate quantum circuits for use in quantum machine learning so that the models are both size-efficient and highly expressive without depending on gradient-based methods, thus overcoming scalability and trainability barriers. The proposed methodology is being adapted to various data types and applications, including tabular data classification, grayscale image classification, multiband image segmentation, and prediction in satellite image sequences. The results obtained are very promising, demonstrating the versatility and robustness of the proposed methodology.