Benchmarking memory and nonlinear processing capacity of quantum recurrent neural networks for time series prediction
Organizations: Galicia Supercomputing Center (CESGA), 15705 Santiago de Compostela, Spain · Computer Graphics and Data Engineering (COGRADE), Departamento de Electrónica e Computación, Universidade de Santiago de Compostela, 15782 Santiago de Compostela, Spain
Abstract
Quantum machine learning has emerged as a promising approach to time series prediction. While quantum reservoir computing has been widely explored for this setting, variational quantum models tailored to sequential data remain comparatively understudied, largely because they are difficult to train. In this work, we present a rigorous benchmark of variational quantum recurrent neural networks (QRNNs) on time series prediction tasks, focusing on two key capabilities: memory and nonlinear processing. We introduce a task-driven methodology that quantifies model capacity, enabling an intuitive and fair comparison across architectures, including classical recurrent baselines. We evaluate the proposed framework on two generic benchmark tasks and on a synthetic dataset describing a chaotic system. Overall, the quantum models achieve performance comparable to their classical counterparts, but classical models scale better with the memory size within the range studied. Notably, one QRNN variant shows higher average capacity on the task designed to jointly probe memory and nonlinear capacity. Moreover, our results show that quantum models offer better capacity when we are able to find a good parameter initialization. However, the increasing sensitivity to initialization leads to a lower probability of finding good solutions as the model size increases. These results provide a principled starting point for designing richer quantum recurrent architectures and for assessing where they may offer practical benefits.
Figures & tables
| Model name | STM [ 33 ] | NARMA [ 34 ] | MG [ 19 ] | Subtotal |
| EMCZ3 | 1,920 | |||
| EMCZ2 [ 51 ] | 1,920 | |||
| EMCX1 | 1,920 | |||
| Hybrid | 1,920 | |||
| VanillaRNN [ 14 ] | 1,920 | |||
| LSTM [ 22 ] | 1,920 |
| Name | Components | Description |
|---|---|---|
| EMCX1 | encode-2P + CX-linear-RA + exp-Z | HEA with real amplitudes |
| EMCZ2 | encode-2P + CZ-linear-2P + exp-Z | HEA with reduced parameters |
| EMCZ3 | encode-3P + CZ-ETOM-3P + exp-Z | HEA with redundant parameters and high entanglement |
| Hybrid | encode-2P + TFIM + full-base | Hybrid (digital + analog) with reservoir-like output layer |
| Model | Fixed | Variable |
|---|---|---|
| EMCX1 | , , | , |
| EMCZ2 | , , | , |
| EMCZ3 | , , | , |
| Hybrid | , , |
Appendix figures & tables6 assets
Supplementary material from the paper’s appendix.
Appendix
| Min. MSE | Min. RMSE | Max. | |
| 1 | 0.0931 | 0.3051 | 0.6914 |
| 2 | 0.0442 | 0.2101 | 0.8537 |
| 3 | 0.0245 | 0.1565 | 0.9192 |
| 6 | 0.0066 | 0.0812 | 0.9789 |
| 20 | 0.0011 | 0.0325 | 0.9965 |
| Min. MSE | Min. RMSE | Max. | |
| 1 | 0.0283 | 0.1681 | 0.9884 |
| 2 | 0.0076 | 0.0869 | 0.8999 |
| 3 | 0.0055 | 0.0742 | 0.9283 |
| 4 | 0.0027 | 0.0521 | 0.9722 |
| 5 | 0.0022 | 0.0467 | 0.9747 |
| 6 | 0.0014 | 0.0372 | 0.9820 |
| maximum knots | 8 |
| degree | 3 |
| penalty | 0.1 |
| penalty | 0.1 |
| bootstrap | 300 |
| Package | Version | Used for |
| TensorFlow | 2.21.0 | CRNN training and inference |
| JAX | 0.9.0.1 | QRNN training and inference |
| Optax | 0.2.7 | QRNN training |
| JiTCDDE | 1.8.3 | Generation of MG dataset |
| Scikit-learn | 1.8.0 | Analysis of results, GAM fits |
| Node | CPU node | GPU node |
|---|---|---|
| 2 Intel Xeon | 2 Intel Xeon | |
| Processor | Ice Lake 8352Y | Ice Lake 8352Y |
| 32 cores 2.2 GHz | 32 cores 2.2 GHz | |
| Memory | 256 GB RAM | 256 GB RAM |
| GPU | None | 2 Nvidia A100 |
Quantum Neural Networks
Quantum neural networks use quantum circuits and measurements to learn representations, make predictions, and model complex data.
See all 148 papers