Article
Physics, Multidisciplinary
Liu Qi, Li Pu, Kai Chao, Hu Chun-Qiang, Cai Qiang, Zhang Jian-Guo, Xu Bing-Jie
Summary: The study introduces a novel method of using time delayed photonic reservoir computing to predict the trajectory of chaotic laser, based on semiconductor lasers as input signal. This approach offers advantages of simple configuration, low training cost, and potential for hardware implementation.
ACTA PHYSICA SINICA
(2021)
Article
Computer Science, Artificial Intelligence
R. Martinez-Pena, J. Nokkala, G. L. Giorgi, R. Zambrini, M. C. Soriano
Summary: This article introduces quantum reservoir computing (QRC) and characterizes its performance, studying the influence of factors such as input injection frequency, time multiplexing, and measured observables on computational capabilities. The study provides optimal input driving conditions and alternative choices for output variables. It establishes a clear understanding of the computational capabilities of a quantum network of spins for reservoir computing, paving the way for future theoretical and experimental research on QRC.
COGNITIVE COMPUTATION
(2023)
Article
Multidisciplinary Sciences
Kazutaka Kanno, Atsushi Uchida
Summary: In this paper, a photonic on-line implementation of reinforcement learning using optoelectronic delay-based reservoir computing is proposed and experimentally validated. The proposed scheme accelerates reinforcement learning process and represents the first hardware implementation of reinforcement learning based on photonic reservoir computing, paving the way for efficient reinforcement learning as a novel photonic accelerator.
SCIENTIFIC REPORTS
(2022)
Article
Chemistry, Multidisciplinary
Yutaro Yamazaki, Kentaro Kinoshita
Summary: This study develops a memristor that can respond to both electrical and optical stimuli, and demonstrates that the timescale of the transient current response of the device can be controlled by applying a small voltage. The computational performance of the device as a physical reservoir is evaluated in an image classification task, showing the potential for optimizing learning accuracy by tuning the device characteristics.
Article
Physics, Applied
Takeshi Yoshimura, Taiki Haga, Norifumi Fujimura, Kensuke Kanda, Isaku Kanno
Summary: In this study, a physical reservoir computing system using a piezoelectric MEMS resonator was demonstrated. Memory characteristics were incorporated through the transient response of the resonator, and a capacitance-resistor series circuit was introduced to improve performance. Benchmark tests showed short-term memory and nonlinearity. The feasibility of MEMS sensors with machine-learning capability was demonstrated.
JAPANESE JOURNAL OF APPLIED PHYSICS
(2023)
Article
Computer Science, Artificial Intelligence
Felix Koester, Serhiy Yanchuk, Kathy Ludge
Summary: This study demonstrates that delay-based reservoir computers can be characterized by a universal master memory function (MMF) and provides linear memory capacity. An analytical description of the MMF is proposed for efficient computing and can be applied to various reservoir scenarios.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2022)
Article
Optics
Jiaoyang Jin, Ning Jiang, Yiqun Zhang, Weizhou Feng, Anke Zhao, Shiqin Liu, Jiafa Peng, Kun Qiu, Qianwu Zhang
Summary: We propose an adaptive time-delayed photonic reservoir computing structure using the Kalman filter algorithm as a training approach. Simulation results on two benchmark tasks demonstrate that the proposed structure with adaptive KF training significantly enhances prediction and equalization performance compared to conventional reservoir computing with least-squares training. Furthermore, introducing a complex mask derived from a enhanced chaotic signal improves the performance further. The work presents a potential way to realize adaptive photonic computing.
Article
Computer Science, Hardware & Architecture
Sosei Ikeda, Hiromitsu Awano, Takashi Sato
Summary: The article introduces a new intermediate representation and hardware-friendly reservoir model for reservoir computing, which successfully classifies multivariate time series data and is compared with traditional models.
IEEE TRANSACTIONS ON COMPUTER-AIDED DESIGN OF INTEGRATED CIRCUITS AND SYSTEMS
(2022)
Article
Nanoscience & Nanotechnology
Ian Bauwens, Krishan Harkhoe, Peter Bienstman, Guy Verschaffelt, Guy Van der Sande
Summary: This study proposes using transfer learning to address the issue of parameter drift in photonic reservoir computing system and reduce the resources required for retraining. Numerical studies on a delay-based system with semiconductor lasers demonstrate that transfer learning can mitigate parameter fluctuations and reduce training requirements for the second task.
Article
Mathematics, Interdisciplinary Applications
Lijun Pei, Mengyu Zhang
Summary: This paper explores the dynamics of a delay-based photonic reservoir computing system with a focus on double Hopf bifurcation. The existence of double Hopf bifurcation points is analyzed and bifurcation diagrams are drawn using DDE-BIFTOOL. Three types of double Hopf bifurcations are found, leading to stable equilibrium, stable periodic, and quasi-periodic solutions in distinct regions. These rich dynamical phenomena can aid in selecting suitable parameter values for optimal performance of the photonic reservoir computing system.
INTERNATIONAL JOURNAL OF BIFURCATION AND CHAOS
(2022)
Article
Computer Science, Artificial Intelligence
Luca Manneschi, Andrew C. Lin, Eleni Vasilaki
Summary: Sparse neural networks are common in machine learning and neuroscience. In this study, the authors introduce sparsity into a reservoir computing network using neuron-specific learnable thresholds of activity. This approach, called SpaRCe, optimizes the sparsity level without affecting network dynamics.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Automation & Control Systems
Yi Sun, Qingjiang Li, Xi Zhu, Cen Liao, Yongzhou Wang, Zhiwei Li, Sen Liu, Hui Xu, Wei Wang
Summary: The study proposes a method of in-sensor computing based on optoelectronic synapses, which allows efficient computation of optical signals. Experimental results demonstrate the effectiveness of the method and its advantage in processing sequential visual information.
ADVANCED INTELLIGENT SYSTEMS
(2023)
Article
Chemistry, Physical
Taro Shingu, Haruki Uchiyama, Takeshi Watanabe, Yutaka Ohno
Summary: Reservoir computing, based on the electrochemical functionalization of carbon nanotube (CNT) thin films, improves the performance of time-series data prediction tasks. By combining CNT electrodes with different functionalization conditions, the dimensionality of the reservoir is enhanced. Moreover, the use of displacement current and redox current improves the dimensionality and memory capacity of the reservoir.
Article
Geosciences, Multidisciplinary
Christopher S. Bretherton
Summary: Physics-informed machine learning is rapidly advancing in geophysical simulation. Recent advances in graph neural networks and vision transformers have shown superior forecasting skills for global weather within 1-7 days, at integration times over 1,000 times faster than conventional models. However, longer simulations deteriorate quickly. Achieving high skill in both weather and climate applications remains a challenging goal for machine learning.
GEOPHYSICAL RESEARCH LETTERS
(2023)
Article
Engineering, Mechanical
Yigong Yang, Pei Zhou, Penghua Mu, Nianqiang Li
Summary: This paper presents the first numerical implementation of photonic reservoir computing based on a spin VCSEL. The proposed system demonstrates fast response and has the potential to achieve high-speed information processing and lower power consumption.
NONLINEAR DYNAMICS
(2022)