Feedback-driven quantum reservoir computing for time-series analysis

TL;DR

This paper introduces a feedback-driven quantum reservoir computing framework that overcomes measurement-induced memory loss, enabling effective time-series analysis and forecasting of quantum signals.

Contribution

It proposes a novel feedback mechanism in quantum reservoir computing that preserves memory and enhances time-series processing capabilities.

Findings

Successfully demonstrates fading-memory property via feedback connections

Identifies three phases depending on feedback strength, with optimal memory at the edge of chaos

Shows improved forecasting of quantum spin system signals

Abstract

Quantum reservoir computing (QRC) is a highly promising computational paradigm that leverages quantum systems as a computational resource for nonlinear information processing. While its application to time-series analysis is eagerly anticipated, prevailing approaches suffer from the collapse of the quantum state upon measurement, resulting in the erasure of temporal input memories. Neither repeated initializations nor weak measurements offer a fundamental solution, as the former escalates the time complexity while the latter restricts the information extraction from the Hilbert space. To address this issue, we propose the feedback-driven QRC framework. This methodology employs projective measurements on all qubits for unrestricted access to the quantum state, with the measurement outcomes subsequently fed back into the reservoir to restore the memory of prior inputs. We demonstrate that our QRC successfully acquires the fading-memory property through the feedback connections, a critical element in time-series processing. Notably, analysis of measurement trajectories reveal three distinct phases depending on the feedback strength, with the memory performance maximized at the edge of chaos. We also evaluate the predictive capabilities of our QRC, demonstrating its suitability for forecasting signals originating from quantum spin systems.