Quantum reservoir computing (QRC) offers an alternative to classical reservoir computing (RC) for machine learning tasks. However, standard designs of QRC cannot reliably perform nonlinear tasks with quantum inputs. This thesis studies a hybrid architecture that combines a quantum reservoir (QR)—modeled as interacting qubits with a transverse field and local disorder—sequentially with an echo state network (ESN), which can faithfully execute nonlinear operations. We demonstrate that this hybrid approach outperforms its standalone components
in both linear and nonlinear tasks, when limited to single-axis measurements, i.e. partial information about the input. The improved performance arises from the information spreading induced by the QR preprocessing layer. These results highlight the potential of hybrid quantum–classical architectures as a promising route for enhanced quantum machine learning.
Supervisors: Roberta Zambrini and Gian Luca Giorgi