Sustainable NARMA-10 Benchmarking for Quantum Reservoir Computing

TL;DR

Problem: determine whether Quantum Reservoir Computing (QRC) can match or exceed classical and hybrid time-series models on the NARMA-10 benchmark while offering improved energy and data efficiency. Approach: a head-to-head benchmarking of QRC, ESN, LSTM, and QLSTM, using fixed quantum reservoirs with trained linear readouts and comprehensive resource metrics, including forecast accuracy via NRMSE and training/inference costs. Contributions: empirical evidence that QRC achieves competitive NRMSE with strong memory efficiency and low trainable parameters, a sustainability-oriented cost reporting framework, and guidance for deploying QRC in data- and compute-constrained settings. Impact: informs the design of energy-conscious quantum-enhanced time-series AI and motivates standardized resource accounting for future scale-up on specialized quantum hardware.

Abstract

This study compares Quantum Reservoir Computing (QRC) with classical models such as Echo State Networks (ESNs) and Long Short-Term Memory networks (LSTMs), as well as hybrid quantum-classical architectures (QLSTM), for the nonlinear autoregressive moving average task (NARMA-10). We evaluate forecasting accuracy (NRMSE), computational cost, and evaluation time. Results show that QRC achieves competitive accuracy while offering potential sustainability advantages, particularly in resource-constrained settings, highlighting its promise for sustainable time-series AI applications.

Figures (2)

• Figure 1: Flow diagram demonstrating the structure of the benchmarks, and summarizing the overall architecture of the paper
• Figure 2: Model comparison focused on sustainability.