arXiv:2610.13694
QTN-RQNN: Quantum Tensor Network-Based Recurrent Quantum Neural Network
Amit S. Patel, Himanshu R. Patel, Bikash K. Behera, Pranav Mehta, Ghanshyam Tejani, Deepak Parashar
Abstract
Resource constraints and imperfect quantum operations remain major challenges for quantum machine learning (QML) on noisy intermediate-scale quantum (NISQ) devices, particularly for fault-diagnosis tasks involving imbalanced and heterogeneous data. This work proposes QTN-RQNN, a Quantum Tensor Network-Based Recurrent Quantum Neural Network that integrates tensor-network-inspired nearest-neighbor interactions with recurrent quantum processing. The model sequentially encodes input features while propagating quantum memory across recurrent steps and incorporates QSMOTE for class balancing. QTN-RQNN is evaluated on four heterogeneous fault-diagnosis datasets SPID, CWRUBD, EFDD, and SFDD and compared with QNN, QTN-QNN, and RQNN baselines through classification, ablation, multi-seed, convergence, and quantum-resource analyses. Results show dataset-dependent predictive behavior, indicating that increased quantum architectural complexity does not universally improve classification. On SPID, QTN-RQNN improves the RQNN F1-score from 0.3077 to 0.4516 and ROC-AUC from 0.4485 to 0.6970, while the remaining datasets exhibit different performance-resource trade-offs and sensitivity to recurrent timesteps and circuit depth. Across all datasets, QTN-RQNN reduces the two-qubit-gate count from 32 to 24 per sample relative to RQNN, corresponding to a 25% reduction without increasing the four-qubit requirement; the same relative reduction is obtained for QTN-QNN over QNN. These findings establish QTN-RQNN as a resource-conscious recurrent quantum architecture and highlight the need to jointly consider predictive performance and circuit-resource requirements for near-term quantum fault diagnosis.
AI Summary
The paper investigates whether nearest-neighbor circuit structure can reduce the resources of recurrent quantum classifiers for imbalanced fault-diagnosis data. QTN-RQNN uses four qubits: two encode successive feature pairs, and two retain quantum memory. A shared cell applies trainable local rotations and nearest-neighbor $R_{ZZ}$ interactions; tracing out the input register defines an input-conditioned channel on memory. Final memory-$Z$ expectations feed a classical sigmoid classifier. Evaluation uses four datasets, training-only QSMOTE balancing, and comparisons with feedforward QNN, structured QTN-QNN, and recurrent RQNN baselines. The default configuration has four timesteps, two layers, and 25 total trainable parameters. Its logical two-qubit-gate count is 24 per sample versus 32 for the ring-entangled RQNN, a 25% reduction, with the same four-qubit requirement but more parameters. Predictive results depend on the dataset. On SPID, QTN-RQNN improves RQNN’s F1 from 0.3077 to 0.4516 and ROC-AUC from 0.4485 to 0.6970, but the feedforward QNN performs better. On CWRUBD, its default F1 is 0.3429 versus RQNN’s 0.5714. On SFDD, ROC-AUC is 0.5093 and every sample is classified positive. Timestep, depth, balancing, and five-seed experiments examine configuration and optimization sensitivity. The paper states that neither recurrence nor structured connectivity provides universal predictive gains. Gate counts are pre-transpilation logical counts, not hardware-native costs; noise-aware and hardware validation remain future work. The timestep ablation also changes feature compression, so it does not isolate recurrence length.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 14 pages, 7 figures