Berlin 2024 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

DY: Fachverband Dynamik und Statistische Physik

DY 49: Focus Session: Computing with Dynamical Systems: New Perspectives on Reservoirs and Applications II – Applications and Quantum RC

DY 49.9: Talk

Thursday, March 21, 2024, 17:45–18:00, BH-N 243

Enhancing the performance of quantum reservoir computing and solving the time-complexity problem by artificial memory restriction — •Saud Cindrak, Kathy Lüdge, and Lina Jaurigue — Institute of Physics, Technische Universität Ilmenau, Weimarer Str. 25, 98693 Ilmenau, Germany

We propose a novel scheme to optimize the performance and reduce the computational cost of quantum reservoir computing. In quantum reservoir computing, a quantum system serves as a reservoir and measurements are performed to obtain the expected values of observables. However, due to the state's collapse after measurement, computations must be repeated multiple times to construct expected values. This becomes challenging for timeseries tasks, where each new input requires the reinitialization of all prior inputs into the system, leading to quadratic time complexity. Another hurdle in reservoir computing lies in tuning nonlinearities. We address these challenges by artificially restricting the reservoir's memory, achieved by reducing the number of reinitialization time steps to a level below the fading memory capacity. With the proposed algorithm, we decrease the time complexity to linear and introduce an experimentally tunable parameter to change the nonlinear response. We demonstrate our approach on both an Ising reservoir and a quantum circuit reservoir and observe an increase in the information processing capacity and a reduction of the prediction errors for the Lorenz timeseries prediction task. [S. Cindrak, B. Donvil, K.Lüdge, L. Jaurigue, ArXiv:2306.12876 (2023). https://doi.org/10.48550/arXiv.2306.12876]

Keywords: Quantum Reservoir Computing; Quantum Machine Learning; Reservoir Computing