We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.LG

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Quantum Physics

Title: Scalable Quantum Neural Networks for Classification

Abstract: Many recent machine learning tasks resort to quantum computing to improve classification accuracy and training efficiency by taking advantage of quantum mechanics, known as quantum machine learning (QML). The variational quantum circuit (VQC) is frequently utilized to build a quantum neural network (QNN), which is a counterpart to the conventional neural network. Due to hardware limitations, however, current quantum devices only allow one to use few qubits to represent data and perform simple quantum computations. The limited quantum resource on a single quantum device degrades the data usage and limits the scale of the quantum circuits, preventing quantum advantage to some extent. To alleviate this constraint, we propose an approach to implementing a scalable quantum neural network (SQNN) by utilizing the quantum resource of multiple small-size quantum devices cooperatively. In an SQNN system, several quantum devices are used as quantum feature extractors, extracting local features from an input instance in parallel, and a quantum device works as a quantum predictor, performing prediction over the local features collected through classical communication channels. The quantum feature extractors in the SQNN system are independent of each other, so one can flexibly use quantum devices of varying sizes, with larger quantum devices extracting more local features. Especially, the SQNN can be performed on a single quantum device in a modular fashion. Our work is exploratory and carried out on a quantum system simulator using the TensorFlow Quantum library. The evaluation conducts a binary classification on the MNIST dataset. It shows that the SQNN model achieves a comparable classification accuracy to a regular QNN model of the same scale. Furthermore, it demonstrates that the SQNN model with more quantum resources can significantly improve classification accuracy.
Comments: 11 pages, 10 figures, accepted by QCE2022
Subjects: Quantum Physics (quant-ph); Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)
Cite as: arXiv:2208.07719 [quant-ph]
  (or arXiv:2208.07719v1 [quant-ph] for this version)

Submission history

From: Qun Li [view email]
[v1] Thu, 4 Aug 2022 20:35:03 GMT (1169kb,D)

Link back to: arXiv, form interface, contact.