QRC-Lab: An Educational Toolbox for Quantum Reservoir Computing

TL;DR

The paper addresses the gap between theoretical quantum reservoir computing and practical, education-oriented workflows by introducing QRC-Lab, a modular, gate-based toolbox for exploring input encoding, reservoir dynamics, measurement, and classical readout under realistic NISQ constraints. It formalizes the QRC framework, defines a configurable pipeline with an encoder, reservoir, and observables, and demonstrates its utility through three educational benchmarks (memory reconstruction, NARMA10 forecasting, and temporal parity). A theory-guided generalization-gap (risk-control) diagnostic ties reservoir capacity to learning performance, guiding systematic ablations and pedagogy. The work emphasizes reproducibility, openness, and pedagogical alignment, providing end-to-end scripts, a Zenodo release, and notebooks to enable classroom use and broad-access research. Overall, QRC-Lab offers a clear, modular path for students and researchers to study memory, nonlinearity, and generalization in quantum temporal processing, with practical implications for designing quantum-enhanced learning systems on NISQ devices.

Abstract

Quantum Reservoir Computing (QRC) has emerged as a strong pa- radigm for Noisy Intermediate-Scale Quantum (NISQ) machine learning, ena- bling the processing of temporal data with minimal training overhead by exploi- ting the high-dimensional dynamics of quantum states. This paper introduces QRC-Lab, an open-source, modular Python framework designed to bridge the gap between theoretical quantum dynamics and applied machine learning work- flows. We provide a rigorous definition of QRC, contrast physical and gate- based approaches, and formalize the reservoir mapping used in the toolbox. QRC-Lab instantiates a configurable gate-based laboratory for studying in- put encoding, reservoir connectivity, and measurement strategies, and validates these concepts through three educational case studies: short-term memory re- construction, temporal parity (XOR), and NARMA10 forecasting as a deliberate stress test. In addition, we include a learning-theory motivated generalization- gap scan to build intuition about capacity control in quantum feature maps. The full source code, experiment scripts, and reproducibility assets are publicly available at: https://doi.org/10.5281/zenodo.18469026.

Figures (4)

• Figure 1: Short-term memory / memory reconstruction task on an ideal backend. The prediction follows the target over part of the test horizon, illustrating fading memory and highlighting where dynamical mixing limits reconstruction of sharp variations.
• Figure 2: NARMA10 nonlinear forecasting on an ideal backend. The prediction captures coarse structure but shows noticeable deviations, illustrating a common failure mode when long memory and nonlinearity are insufficiently expressed by the reservoir/observable configuration.
• Figure 3: Temporal parity (XOR) task on an ideal backend. The near-perfect overlap indicates that the quantum reservoir generates features that make a nonlinearly separable temporal rule linearly decodable by the readout.
• Figure 4: Generalization-gap scan in QRC-Lab: training and test $R^2$ as a function of the number of qubits, highlighting diminishing returns and a regime of potential overfitting.