Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Quantum Extreme Reservoir Computing for Phase Classification of Polymer Alloy Microstructures

Arisa Ikeda, Akitada Sakurai, Kae Nemoto, Mayu Muramatsu

TL;DR

This work investigates applying quantum extreme reservoir computing (QERC) to phase classification of SCFT-generated polymer alloy microstructures, aiming to bridge quantum learning with materials informatics. By compressing high-dimensional image data with PCA, encoding into a quantum reservoir built from Clifford+T circuits, and classifying via a downstream neural network, the approach achieves high-accuracy phase discrimination with around seven qubits. Phase diagrams reconstructed from QERC outputs offer interpretable mappings between quantum model behavior and material phase boundaries, and analyses of encoder design reveal practical guidelines for generalization and robustness. The results establish a realistic benchmark for quantum learning on engineering datasets and highlight the potential of QERC for industrial materials informatics.

Abstract

Quantum machine learning (QML) is expected to offer new opportunities to process high-dimensional data efficiently by exploiting the exponentially large state space of quantum systems. In this work, we apply quantum extreme reservoir computing (QERC) to the classification of microstructure images of polymer alloys generated using self-consistent field theory (SCFT). While previous QML efforts have primarily focused on benchmark datasets such as MNIST, our work demonstrates the applicability of QERC to engineering data with direct materials relevance. Through numerical experiments, we examine the influence of key computational parameters-including the number of qubits, sampling cost (the number of measurement shots), and reservoir configuration-on classification performance. The resulting phase classifications are depicted as phase diagrams that illustrate the phase transitions in polymer morphology, establishing an understandable connection between quantum model outputs and material behavior. These results illustrate QERC performance on realistic materials datasets and suggest practical guidelines for quantum encoder design and model generalization. This work establishes a foundation for integrating quantum learning techniques into materials informatics.

Quantum Extreme Reservoir Computing for Phase Classification of Polymer Alloy Microstructures

TL;DR

This work investigates applying quantum extreme reservoir computing (QERC) to phase classification of SCFT-generated polymer alloy microstructures, aiming to bridge quantum learning with materials informatics. By compressing high-dimensional image data with PCA, encoding into a quantum reservoir built from Clifford+T circuits, and classifying via a downstream neural network, the approach achieves high-accuracy phase discrimination with around seven qubits. Phase diagrams reconstructed from QERC outputs offer interpretable mappings between quantum model behavior and material phase boundaries, and analyses of encoder design reveal practical guidelines for generalization and robustness. The results establish a realistic benchmark for quantum learning on engineering datasets and highlight the potential of QERC for industrial materials informatics.

Abstract

Quantum machine learning (QML) is expected to offer new opportunities to process high-dimensional data efficiently by exploiting the exponentially large state space of quantum systems. In this work, we apply quantum extreme reservoir computing (QERC) to the classification of microstructure images of polymer alloys generated using self-consistent field theory (SCFT). While previous QML efforts have primarily focused on benchmark datasets such as MNIST, our work demonstrates the applicability of QERC to engineering data with direct materials relevance. Through numerical experiments, we examine the influence of key computational parameters-including the number of qubits, sampling cost (the number of measurement shots), and reservoir configuration-on classification performance. The resulting phase classifications are depicted as phase diagrams that illustrate the phase transitions in polymer morphology, establishing an understandable connection between quantum model outputs and material behavior. These results illustrate QERC performance on realistic materials datasets and suggest practical guidelines for quantum encoder design and model generalization. This work establishes a foundation for integrating quantum learning techniques into materials informatics.
Paper Structure (14 sections, 6 equations, 6 figures, 1 table)

This paper contains 14 sections, 6 equations, 6 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Classification setup and dataset creation
  4. Visualization of phase diagrams based on classification results
  5. Quantum Extreme Reservoir Computing (QERC)
  6. Results
  7. Classification accuracy vs the number of qubits
  8. Verification of the reservoir parts using only random Clifford circuits or only T-gates
  9. Discussions
  10. Verification of factors contributing to accuracy dependence on the number of qubits
  11. Evaluation of generalization performance of QML models on parameter space
  12. Conclusion
  13. Preparing polymer alloy microstructure images using SCFT
  14. Labeling of microstructures

Figures (6)

  • Figure 1: Phase diagram for microstructures of polymer alloys and conceptual diagram of QERC. (a) Schematically represents the differences in microstructure due to the interaction $f$ between segments and the product of the $\chi$ parameter and the degree of polymerization $N$. (b) shows an example of a phase diagram based on the microstructure of a polymer alloy, (c) indicates the correct labels in the phase diagram. (d) is a conceptual diagram of QERC.
  • Figure 2: Visualization of the phase diagram based on classification results: For each grid point in the parameter space, ten test images were generated using different random seeds. The class label for each grid point was determined by majority voting on the QERC classification results, and the corresponding color was assigned accordingly. The transparency was adjusted in proportion to the majority ratio.
  • Figure 3: Difference in accuracy based on qubit count. Solid lines show results using QERC, dashed lines show results using a single-layer neural network (NN) without a quantum reservoir. Since random number effects significantly impact linear classification, results vary easily even without changing conditions; thus, results were verified three times each.
  • Figure 4: Visualization of phase diagrams based on QERC predictions: (a) The number of qubits is 5 to 8. The number of shots is 2,048. The reservoir is random Clifford + T-layer. (b) The reservoir is random Clifford or T-layer. The number of qubits is 8 and the number of shots is 2,048.
  • Figure 5: Visualization of contribution rates and cumulative contribution rates after PCA application and phase diagrams: (a) PCA contribution rate and cumulative contribution rate. (b) Phase diagram : (b-1) 7 qubits result, encoding components 1 to 12 and components 15 to 16; (b-2) 8 qubits result, encoding components 1 to 12 and components 17 to 20.
  • ...and 1 more figures