Mimicking an Information Reservoir by Superconducting Quantum Circuits

Korkmaz, Ufuk; Sanga, Cem; Türkpençe, Deniz

doi:10.1109/ismsit52890.2021.9604610

Cited by 5 publications

(1 citation statement)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Quantum reservoirs have been experimentally realized with a variety of quantum computing technologies including superconducting circuits [8,18], quantum photonic memristors [19], NMR [20], trapped ions [21], Rydberg atoms [22], and quantum spin systems [23]. Qudit systems have been implemented [24].…”

Section: Introductionmentioning

confidence: 99%

Optimizing Quantum Noise-induced Reservoir Computing for Nonlinear and Chaotic Time Series Prediction

Fry¹,

Deshmukh²,

Chen³

et al. 2023

Preprint

View full text Add to dashboard Cite

Quantum reservoir computing (QRC) has been strongly emerging as a time series prediction approach in quantum machine learning (QML). This work is the first to apply optimization to resource noise in a quantum reservoir to effectively improve time series prediction. Based on this development, we propose a new approach to quantum circuit optimization.We build on the work of Suzuki et al., who used quantum hardware noise as an essential resource in quantum noise-induced reservoir (QNIR) computer for generating non-trivial output sequences, and we achieve a novel, optimized QNIR, in which the artificial noise channels are parameterized. To optimize the parameterized resource noise, we use dual annealing and evolutionary optimization. Another essential component of our approach is reducing quantum resources in the number of artificial noise models, number of qubits, entanglement scheme complexity, and circuit depth. A key result is the reduction from a general multi-component noise model to a single reset noise model.Reservoir computers are especially well-suited for modelling nonlinear dynamical systems. In this paper we consider NARMA and Mackey-Glass systems, which are common univariate time series benchmarks for reservoir computers and recurrent neural networks. Recently QRCs have demonstrated good prediction capability with small numbers of qubits. QNIR simulations based on our optimization approach demonstrate high performance on NARMA benchmarks while using only a 12-qubit reservoir and a single noise model. Good prediction performances over 100 timesteps ahead for the Mackey-Glass system are demonstrated in the chaotic regime. In addition, these results provide valuable insight into resource noise requirements for the QNIR.

show abstract