Speaker
Description
Quantum Reservoir Computing (QRC) is a quantum machine-learning framework that exploits the dynamics of quantum systems as a fixed reservoir to process time-dependent input signals, with learning confined to a classical readout layer. This makes QRC efficient to train and well-suited for real-time prediction [Fuji2017, Nakajima2021, Pena2023] . In this work, we investigate a QRC framework based on an Ising model reservoir, studying the effect of noise on the accuracy of predictions. Performance is evaluated on standard benchmarks of short-term memory (STM) and Mackey-Glass time-series prediction tasks. Our results demonstrate strong predictive performance on complex time series, with the number of qubits required by a quantum reservoir being orders of magnitude smaller than the number of nodes required in classical reservoirs. We find that a fine-tuned Ising network of 5-10 qubits can match the predictive performance of classical reservoirs with 100-500 nodes on the chaotic Mackey-Glass series, consistent with previous reports [Fuji2017]. The number of readout features, constructed from k-body observables, such as single (Z) and two-qubit (ZZ) observables, is comparable to that required by classical reservoir models. By comparing two operating regimes of QRC, we elucidate the role of noise induced by environmental processes. Rather than being purely detrimental, noise can stabilise reservoir dynamics and enhance memory capacity [Fuji2017]. We identify specific noise regimes in which QRC achieves memory capacities exceeding those of comparable noiseless models while maintaining performance. This allows us to pinpoint optimal noise rates, demonstrating QRC as a hardware-compatible framework for time-series prediction in forecasting and monitoring tasks.
We acknowledge the support of the 2025 AQSN Microgrant scheme sponsored by the Hon Hai Research.
| I am the presenting author | Yes |
|---|