Hybrid Quantum-Classical CMB Pattern Classification Using Structured Feature Compression

This paper presented a hybrid quantum-classical frame- work for classifying CMB temperature-pattern patches derived from a public WMAP sky map. The workflow converted the original map into a 100,000-sample benchmark of length-9 local patches and assigned cold, neutral, and hot labels using percentile-based thresholds. Classical baselines showed that the dataset is highly separable: logistic regression achieved 99.955% accuracy, SVM achieved 99.575%, and random forest achieved 97.250%. A Ramanujan-inspired compression stage then reduced the feature space from 9 dimensions to 4 structured descriptors while still preserving 99.880% accuracy for compressed logistic regression. The quantum branch used a 4-qubit, 3-layer variational classifier on a balanced binary subset of 400 training and 120 test samples. After angle encoding and 25 epochs of opti- mization, the best quantum configuration achieved 99.167% accuracy, 0.99 precision, 0.99 recall, and 0.99 F1-score, with confusion values of 59 true negatives, 1 false positive, 0 false negatives, and 60 true positives. The quantum loss also decreased from 0.4629 to 0.0894, showing stable optimization behavior. These findings demonstrate that low-qubit variational classification can remain highly competitive after structured feature compression. However, the study does not establish quantum advantage, because the strongest classical baselines remain equal or stronger on both the full three-class task and the fair binary comparison. The most appropriate interpretation is that the project demonstrates the feasibility, interpretability, and reproducibility of hybrid QML for cosmology-inspired data. Future work should investigate richer CMB represen- tations and noisy-hardware validation. It also supports future hardware-aware quantum cosmology experiments.

[1]     J.-C. Yang, S. Zhang, and C.-X. Yue, “A quantum machine learning classifier to search for new physics,” Journal of High Energy Physics, vol. 2026, no. 23, 2026, doi: 10.1007/JHEP01(2026)023. [2]     S. Jabeen, D. Kurdydyk, A. Palnitkar, et al., “Quantum machine learning for state tomography using classical data,” Quantum Machine Intelli- gence, vol. 8, art. no. 40, 2026, doi: 10.1007/s42484-026-00345-0. [3]     F. A. Alotaibi, M. S. N. Yagmahan, K. A. Alobaid, et al., “Quantum inspired feature engineering for explainable EEG signal classification,” Scientific Reports, vol. 16, 2026, doi: 10.1038/s41598-026-41821-8. [4]     R. K. Jha, N. Kasabov, S. Bhattacharyya, D. Coyle, and G. Prasad, “Comparative performance analysis of quantum feature maps for quan- tum kernel-based machine learning,” Scientific Reports, vol. 16, art. no. 8142, 2026, doi: 10.1038/s41598-026-39392-9. [5]     J. A. Bravo-Montes, A. Mart´ın-Toledano, C. Velasco-Gallego, et al., “Design of a hybrid quantum machine learning architecture and analysis of quantum noise effects,” Scientific Reports, 2026, doi: 10.1038/s41598- 026-42216-5. [6]     N. Dowling, M. T. West, A. Southwell, et al., “Adversarial robustness guarantees for quantum classifiers,” npj Quantum Information, vol. 12, art. no. 16, 2026, doi: 10.1038/s41534-025-01129-3. [7]     I. Montalba´n, E. F. Combarro, and J. Ranilla, “A framework to study the relationship between classical data and quantum characteristics in quantum machine learning,” The Journal of Supercomputing, vol. 82, art. no. 234, 2026, doi: 10.1007/s11227-026-08396-7. [8]     G. Singh, H. Jin, and K. M. Merz Jr., “Benchmarking MedMNIST dataset on real quantum hardware,” Scientific Reports, vol. 16, art. no. 9017, 2026, doi: 10.1038/s41598-026-35605-3. [9]     V. Lopatkin, A. Sagingalieva, L. Lusnig, et al., “Quantum hybrid feature selector,” EPJ Quantum Technology, 2026, doi: 10.1140/epjqt/s40507- 026-00491-1. G. S. Franco, F. Mahlow, P. M. Prado, et al., “Quantum phases classification using quantum machine learning with SHAP-driven feature selection,” Quantum Information Processing