arXiv:2610.13388
Quantum extreme learning for forecasting experimental chaotic electronic circuits
Mohammad Sharifian, Qingyu Li, Chiranjib Mukhopadhyay, Abolfazl Bayat, Ludovico Minati
Abstract
Quantum reservoir computing and its special case, quantum extreme learning, are powerful tools for predicting classical discrete time-series with quantum devices. Forecasting continuous signals, however, requires careful discretization as oversampling artificially boosts short-term accuracy but fails over realistic intervals, while undersampling captures too few features and yields poor predictions. Here, we experimentally generate four continuous time-series signals from a chaotic electronic circuit for forecasting with quantum extreme learning. Since simplified circuit models do not fully capture real behaviour, it is unclear whether algorithms that predict simulated time-series also work for experimental signals. We address discretization via two approaches: (i) sampling local peaks/troughs; and (ii) uniform sampling at the autocorrelation timescale. The former captures resonant behaviour, while the latter fully captures chaotic features as well as allowing the reconstruction of chaotic attractors. Our algorithm achieves high forecasting accuracy for both simulated and experimental signals, demonstrating its power for continuous signal processing on near-term quantum simulators, and remains robust against moderate decoherence.
AI Summary
The paper investigates whether quantum extreme learning (QEL) can forecast experimentally measured continuous chaotic signals, and how sampling should be chosen to avoid trivially accurate predictions from oversampling. It studies two voltages and two currents from an autonomous transistor-based chaotic oscillator, alongside signals generated by a reduced four-variable differential-equation model. The quantum predictor is classically simulated with up to 11 qubits. Past samples are rotation-encoded, evolved once under a random nearest-neighbour XYZ Hamiltonian, and mapped from local Pauli expectation values to predictions by ridge regression. Two discretizations are compared: successive extrema and uniformly spaced samples, with spacing determined by the first $1/e$ autocorrelation crossing. Selected lags span 7–68 acquisition samples, each $3.2\,\mathrm{ns}$. Training uses $5$–$210\,\mu\mathrm{s}$ and testing $210$–$300\,\mu\mathrm{s}$; reservoir configurations are assessed over 100 Hamiltonian realizations. For uniform sampling with all four signals as inputs, the best configurations all use 11 qubits. Standardized test RMSE ranges from $0.14$ to $0.53$ for simulated signals and $0.17$ to $0.50$ for experimental signals. Small LSTMs with 30–112 trainable parameters match or outperform these results. Predictions use measured past inputs rather than autonomous recursive rollout. Phase-space plots show partial recovery of attractor geometry, with displaced or distorted excursions. Lindblad simulations examine amplitude damping and small training–testing rate variations. Stated limitations include the small simulated quantum reservoirs, the circuit model’s omitted parasitic dynamics and measurement effects, and incomplete attractor recovery. Quantum-hardware forecasting is not demonstrated.
Error found: The abstract's claim that autocorrelation-timescale sampling fully captures chaotic features exceeds the reported partial, visibly distorted attractor reconstruction. The autocorrelation lag spaces historical inputs and determines the forecast horizon, but predictions still advance at the original acquisition interval, so reduced processing throughput is not demonstrated. The supplied amplitude-damping analysis lacks plotted results and numerical mean-rate ranges needed to assess the claim of robustness to moderate decoherence. Reservoir configurations and illustrative Hamiltonian realizations are selected using testing-interval RMSE without a separate final evaluation, making their reported performance optimistically selected. Attractor plots constructed from predictions continually supplied with actual historical observations do not independently demonstrate autonomous chaotic dynamics or generalization beyond the selected evaluation interval.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.