We are thrilled to announce that our latest paper, “Hierarchical Quantum Circuit Representations for Neural Architecture Search,” has just been published in the esteemed npj Quantum Information.
Our work introduces an exciting paradigm by borrowing techniques from the field of Neural Architecture Search (NAS). In classical machine learning, NAS has automated neural network architecture design and achieved state-of-the-art performance. We propose to extend these concepts into the quantum realm.
This paper presents a framework for representing quantum circuit architectures, allowing for design and architecture search. The magic of our approach lies in its modularity, adaptability, and ability to reveal repeating patterns, which mirror the common features in constructing neural and tensor networks.
At the heart of our study, we demonstrate the crucial role of circuit architecture in quantum machine learning. We create a family of Quantum Convolutional Neural Networks (QCNNs) and evaluate them on a music genre classification dataset, GTZAN. Our findings underscore the potential and versatility of QCNNs and quantum machine learning as a whole.
But we didn’t stop there. We went a step further by employing a genetic algorithm to perform Quantum Phase Recognition (QPR) as an example of architecture search with our representation. This approach demonstrates the effectiveness of our representation in practical applications, providing a promising starting point for further exploration in quantum machine learning architectures.
To make our work accessible to everyone and encourage further exploration, we have also developed and released an open-source Python package. This package facilitates dynamic circuit creation and circuit search space design, enabling others to experiment with NAS in quantum circuits.
We are proud to contribute to this growing field and are excited to see where these advancements will take us next. We invite you to read our full paper to delve into the details of our research: Hierarchical quantum circuit representations for neural architecture search.
Matt Lourens has written a notice blog post introducing the paper’s main results. You can find the corresponding software package on GitHub.
Stay tuned for more exciting updates from our group!