Information Processing Capacity of Spin-Based Quantum Reservoir Computing Systems
- PDF / 2,155,955 Bytes
- 12 Pages / 595.224 x 790.955 pts Page_size
- 53 Downloads / 247 Views
Information Processing Capacity of Spin-Based Quantum Reservoir Computing Systems ˜ 1 R. Mart´ınez-Pena
· J. Nokkala1 · G. L. Giorgi1 · R. Zambrini1 · M. C. Soriano1
Received: 3 March 2020 / Accepted: 15 September 2020 © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract The dynamical behavior of complex quantum systems can be harnessed for information processing. With this aim, quantum reservoir computing (QRC) with Ising spin networks was recently introduced as a quantum version of classical reservoir computing. In turn, reservoir computing is a neuro-inspired machine learning technique that consists in exploiting dynamical systems to solve nonlinear and temporal tasks. We characterize the performance of the spin-based QRC model with the Information Processing Capacity (IPC), which allows to quantify the computational capabilities of a dynamical system beyond specific tasks. The influence on the IPC of the input injection frequency, time multiplexing, and different measured observables encompassing local spin measurements as well as correlations is addressed. We find conditions for an optimum input driving and provide different alternatives for the choice of the output variables used for the readout. This work establishes a clear picture of the computational capabilities of a quantum network of spins for reservoir computing. Our results pave the way to future research on QRC both from the theoretical and experimental points of view. Keywords Machine learning · Quantum reservoir computing · Information Processing Capacity
Introduction Machine learning has become one of the fastest-growing research lines in the last years, with deep learning being a prominent example [1]. Applied to many fields like computer vision [2], physical sciences [3], medicine [4] or language processing [5], machine learning techniques enable us to solve problems that were very hard or even impossible to tackle with more traditional tools. A specific group of problems that belongs to this category are the ones that involve the processing of temporal signals, such as speech recognition [6], time series prediction [7] or channel equalization [8].
This article belongs to the Topical Collection: Trends in Reservoir Computing Guest Editors: Claudio Gallicchio, Alessio Micheli, Simone Scardapane, Miguel C. Soriano R. Mart´ınez-Pe˜na
[email protected] 1
Instituto de F´ısica Interdisciplinar y Sistemas Complejos (IFISC, UIB-CSIC), Campus Universitat de les Illes Balears, E-07122, Palma de Mallorca, Spain
The general concept of Recurrent Neural Network (RNN) encloses one of the main techniques employed nowadays to solve temporal tasks [9]. A RNN is a neural network whose nodes are recursively connected, which allows information to remain in the network through time, giving the system “memory.” Among all the RNN techniques, reservoir computing (RC) is a promising line of research that exploits dynamical systems to process the input information [10, 11]. The inception of this field comes from two different
Data Loading...
