Nonlinear Theory and Its Applications, IEICE
Online ISSN : 2185-4106
ISSN-L : 2185-4106
Current issue
Displaying 1-13 of 13 articles from this issue
Special Section on Cellular Dynamical Systems
  • Hiroyuki Torikai
    Article type: FOREWORD
    2024 Volume 15 Issue 1 Pages 1
    Published: 2024
    Released on J-STAGE: January 01, 2024
    JOURNAL OPEN ACCESS
    Download PDF (89K)
  • Jun Shibayama
    Article type: Paper
    2024 Volume 15 Issue 1 Pages 2-16
    Published: 2024
    Released on J-STAGE: January 01, 2024
    JOURNAL OPEN ACCESS

    This paper introduces the finite-difference time-domain (FDTD) method widely used for the analysis of electromagnetic problems. The original FDTD method is based an explicit formulation in time, allowing simple arithmetic operations without calculating simultaneous equations. However, at the cost of such a simple calculation, the FDTD method has a limitation on the choice of the time step size, known as the Courant-Friedrichs-Lewy (CFL) condition. To remove the CFL condition, the locally one-dimensional (LOD) scheme has been applied to the implicit FDTD formulation. Here, we review the formulation of the FDTD method and describe its application to implicit calculations, particularly with the use of the LOD scheme.

    Download PDF (373K)
  • Megumi Akai-Kasaya, Kento Igarashi, Tetsuya Asai
    Article type: Paper
    2024 Volume 15 Issue 1 Pages 17-35
    Published: 2024
    Released on J-STAGE: January 01, 2024
    JOURNAL OPEN ACCESS

    Molecular neuromorphic devices are composed of random and extremely dense networks of single-walled carbon nanotubes (SWNTs) complexed with polyoxometalate (POM). Such devices are expected to possess the rudimentary ability of reservoir computing (RC), which utilizes the signal-response dynamics and a certain degree of network complexity. The potential of non-linear dynamic systems to serve as reservoirs has attracted considerable attention for the physical realization of RC. In this study, three theoretical approaches are introduced for the physical component of a reservoir with dynamic response and nonlinearity. The first is the cellar automata model for the random network of the model, second is a hardware system working as a reservoir with one simple form of nonlinearity that reflects the intrinsic characteristics of the materials, and third is a large simulation model that includes the negative differential resistance of the POM. Although the two cell-ligation models incorporated different molecular properties, they both exhibited excellent reservoir properties. Furthermore, a simple nonlinear system was driven as a reservoir and demonstrated excellent performance in speech recognition and other functions. Our results are expected to facilitate the development of material-based RC devices.

    Download PDF (3518K)
  • Hidetaka Marumo, Takashi Matsubara
    Article type: Paper
    2024 Volume 15 Issue 1 Pages 36-53
    Published: 2024
    Released on J-STAGE: January 01, 2024
    JOURNAL OPEN ACCESS

    Accurate understanding of the environment is crucial for autonomous driving and robot automation. Depth sensors, including light detection and ranging and depth cameras, are attracting attention. It is practical to treat the depth information in a depth image form. With the progress in Artificial Intelligence, many deep neural networks have been proposed for the segmentation of depth images. However, no method has focused on the difference in scale within an image caused by a 3-dimensional to 2-dimensional projection. We proposed a new scale-equivariant convolution method that focuses on the relationship between the object distance and scale ratio in the image.

    Download PDF (7069K)
  • Masatoshi Sato, Hisashi Aomori, Tsuyoshi Otake
    Article type: Paper
    2024 Volume 15 Issue 1 Pages 54-71
    Published: 2024
    Released on J-STAGE: January 01, 2024
    JOURNAL OPEN ACCESS

    In this paper, we propose automated and accelerated Graph Cut based image segmentation utilizing U-Net. In Graph Cut, seeded image generation is an important element for obtaining highly accurate output images. By utilizing U-Net to automate the generation of seed images, which until now has been done manually, a highly accurate and accelerated Graph Cut are realized. In the simulation, the U-Net is trained from only one original image to generate seeded images for Graph Cut. Using that seeded images, we evaluated the accuracy of test images in which the object and background were segmented by Graph Cut.

    Download PDF (11201K)
  • Nobuyuki Hirami, Takeshi Kamio, Hisato Fujisaka
    Article type: Paper
    2024 Volume 15 Issue 1 Pages 72-86
    Published: 2024
    Released on J-STAGE: January 01, 2024
    JOURNAL OPEN ACCESS

    Three types of networks, neighbor-coupled, all-coupled, and cross-coupled networks, are analyzed numerically or analytically. Inverter cells organizing these networks are built of quantum effect devices. In this paper, single-electron tunneling junctions are adopted to build the cells and their probability distribution of switching delay is approximated to exponential distribution. Although neighbor-coupled networks are simple in structure, their wave propagation mode depends on the scale of the network, which may be a disadvantage in the application to clock delivery. All-coupled networks take only a single-mode. However, period of wave circulation on the networks depends on network scale. Nearest-neighbor cross-coupling is found to be a technique to mitigate the problems of the former two types of networks.

    Download PDF (694K)
  • Akiyoshi Yasuda, Takeshi Kamio, Ibuki Nakamura, Hisato Fujisaka
    Article type: Paper
    2024 Volume 15 Issue 1 Pages 87-106
    Published: 2024
    Released on J-STAGE: January 01, 2024
    JOURNAL OPEN ACCESS

    An error correction network that consists of multi-tone delta-sigma (DS) domain mix-select cells connected each other with bit-transfer lines is proposed. The network functions as a factor graph executing min-max belief propagation algorithm with adders (mixers) and min/max selectors operating on multi-tone DS modulated signals and outputting signals in the same form. The network is used repeatedly q/p times for error correction of all decomposed 2p-QAM sub-symbols of received 2q-QAM symbols, p<q. Then, the network is considered as a time-sharing processor for the error correction of 2q-QAM symbols. Each cell of the network processes all elements of several 2p-dimensional message vectors and passes them to other cells. The one message vector is modulated to one 2p-tone DS modulated sequence and the one cell processes the 2p elements of the vector in turn. Then, each cell is considered as a time-sharing processor of the message vector. The former time-sharing error correction and the latter time-sharing message processing are executed in slow and fast time-scales, respectively. Thus, the network is a dual-scale time-sharing system. The network has two advantages. First, the network can be adopted to adaptive QAM communication. Even if modulation order 2q is changed to 2q′, the same network operates q′/p times with nothing changed. Secondly, elements of a message vector decrease from 2q to (q/p)2p. From results of circuit simulation, it was found that the circulated DS modulation noise in the network did not impair the error correction ability when the bit-transfer lines were long enough.

    Download PDF (4039K)
  • Naruki Hagiwara, Takafumi Kunimi, Kota Ando, Megumi Akai-Kasaya, Tetsu ...
    Article type: Paper
    2024 Volume 15 Issue 1 Pages 107-118
    Published: 2024
    Released on J-STAGE: January 01, 2024
    JOURNAL OPEN ACCESS

    Predictive coding (PC) based on the free-energy principle (FEP) has been extensively explored for the next-generations artificial intelligence. In this study, we constructed PC network models that can implement perception and unsupervised learning by minimizing variational free energy through neural dynamics. Furthermore, these models were applied to practical tasks, such as image discrimination and real-time prediction. For implementation as neuromorphic hardware with biological plausibility, the performance of PC networks was evaluated by applying techniques used in recent neuromorphic engineering, such as augmented direct feedback alignment and physical reservoir systems. These results pave the way for neuromorphic hardware capable of autonomous perception and learning based on the FEP.

    Download PDF (9098K)
  • Hideharu Toda, Shuichi Tajima, Kazuki Nakashima, Tsuyoshi Otake, Hisas ...
    Article type: Paper
    2024 Volume 15 Issue 1 Pages 119-131
    Published: 2024
    Released on J-STAGE: January 01, 2024
    JOURNAL OPEN ACCESS

    This paper proposes the hierarchical lossless color image coding method using CeNNs (cellular neural networks) based predictors. CeNNs are inherently only processing grayscale images, although color image compression utilizes correlations within the RGB color space. To deal with this problem, YCoCg-R color space with low color correlation is employed. The histogram packing technique is also introduced to suppress the expansion of the dynamic range of the chroma. Experimental results confirmed that the proposed method has better coding performance than the conventional method. Compared to FLIF (free lossless image format), the proposed method reduces the bit rate by 8.2%.

    Download PDF (9702K)
  • Kentaro Takeda, Masato Ishikawa, Hiroyuki Torikai
    Article type: Paper
    2024 Volume 15 Issue 1 Pages 132-152
    Published: 2024
    Released on J-STAGE: January 01, 2024
    JOURNAL OPEN ACCESS

    A novel membrane potential model whose nonlinear dynamics is described by an ergodic discrete difference equation is presented. It is shown that the model can exhibit various neuron-like behaviors depending on parameter values. Using the ergodic membrane potential model, a novel multi-compartment neuron model (soma-dendrite-synapse model) is presented. It is shown that the model can exhibit various dendritic phenomena depending on parameter values. Based on detailed analyses of the dendritic phenomena, a design procedure of the multi-compartment neuron model to realize conditioning functions is proposed. Furthermore, the neuron model is implemented by a field programmable gate array and experiments validate its conditioning functions. It is then shown that the proposed neuron model consumes much fewer hardware resources and much lower power than a typical conventional multi-compartment neuron model.

    Download PDF (2455K)
Regular Section
  • Yusuke Matsuoka
    Article type: Paper
    2024 Volume 15 Issue 1 Pages 153-167
    Published: 2024
    Released on J-STAGE: January 01, 2024
    JOURNAL OPEN ACCESS

    This paper presents the implementation of an analog-to-digital (A/D) converter (ADC) with a window using a field programmable analog array (FPAA). This ADC exhibits superstable behavior that corresponds to the digital output sequence and realizes rate encoding A/D conversion. To implement the ADC using an FPAA, a scaling adjustment for the mathematical model that describes the dynamics is performed. A hardware experiment confirms that the dynamics of the ADC are realized by the circuit implementation of the FPAA. The dynamics of the ADC in the circuit implementation with the analog input as an external input are confirmed.

    Download PDF (8201K)
  • Haruna Matsushita, Tomoki Gotoh, Takuji Kousaka
    Article type: Paper
    2024 Volume 15 Issue 1 Pages 168-182
    Published: 2024
    Released on J-STAGE: January 01, 2024
    JOURNAL OPEN ACCESS

    Nested-layer particle swarm optimization (NLPSO) is a powerful bifurcation parameter detection method that requires no careful initialization, second-order partial derivative, or Lyapunov exponents. Previous studies have proven the effectiveness of NLPSO for codimension-one bifurcation, but not for codimension-two bifurcation phenomena. This study proposes a novel objective function for NLPSO for detecting six types of bifurcations, including three codimension-one bifurcations and three codimension-two bifurcations. By setting three parameters according to predetermined laws in each target bifurcation parameter, the modified NLPSO with the proposed objective function accurately detected both codimension-one and codimension-two bifurcation parameters in discrete-time dynamical systems and non-autonomous systems, without a change in the search algorithm. Furthermore, especially for detecting bifurcation parameters located at the ends of the Neimark-Sacker bifurcation curve, the modified NLPSO with the proposed objective function significantly reduced the computational amount using different codimension-one and codimension-two bifurcation searches.

    Download PDF (263K)
  • Junya Kato, Gouhei Tanaka, Ryosho Nakane, Akira Hirose
    Article type: Paper
    2024 Volume 15 Issue 1 Pages 183-204
    Published: 2024
    Released on J-STAGE: January 01, 2024
    JOURNAL OPEN ACCESS

    We propose reconstructive reservoir computing (RRC) which performs better anomaly detection in time-series signals than forecasting-based methods. In this paper, reconstruction means that a neural network generates past input signals. RRC reconstructs a past normal signal for anomaly detection using an echo state network which can learn quickly and stably. We expect that it is easier to restore a past normal signal than to predict an unknown future normal signal. For anomaly detection, we compute an instantaneous reconstruction error. The reconstruction error larger than a threshold is a sign of anomaly. We conduct experiments using sound data obtained from a pump. In the experiments, we pay attention to a time lag between input and output to be reconstructed since we assume that an excessive time lag makes reconstruction difficult due to signal attenuation in the network. Experimental results show that if the time lag is moderate, the reconstruction error of the normal signal is lower than the forecasting error of the same signal. Furthermore, we show that RRC with the appropriate time lag has a better anomaly detection performance index than forecasting-based methods.

    Download PDF (2583K)
feedback
Top