Delays can have a negative impact on the stability of various systems. However, it was reported that the transmission delays of signals among oscillators can stabilize unstable equilibrium points in coupled oscillators, which is called amplitude death (AD). AD caused by delays has been extensively researched in the field of nonlinear science and has attracted attention for various engineering applications. In this article, we review the latest studies on AD caused by delays in oscillator networks, in particular those that considered heterogeneous delays and time-varying networks.

Nonlinear phenomena can be analyzed via linear techniques using operator-theoretic approaches. Data-driven method called the extended dynamic mode decomposition (EDMD) and its variants, which approximate the Koopman operator associated with the nonlinear phenomena, have been rapidly developing by incorporating machine learning methods. Neural ordinary differential equations (NODEs), which are a neural network equipped with a continuum of layers, and have high parameter and memory efficiencies, have been proposed. In this paper, we propose an algorithm to perform EDMD using NODEs. NODEs are used to find a parameter-efficient dictionary which provides a good finite-dimensional approximation of the Koopman operator. We show the superiority of the parameter efficiency of the proposed method through numerical experiments.

A chaotic neural network reservoir (CNNR), in which a chaotic neural network model is used as a reservoir layer, efficiently introduces chaotic dynamics without violating the echo state property criterion. However, it is unclear which aspect of the complex reservoir dynamics contributes to the performance of CNNR. In this study, we analyze the dynamics of CNNR, which is used to engage time-series prediction tasks, and focus particularly on the design and evaluation of CNNR hardware. First, the memory capacity of CNNR is estimated using the NARMA-30 time series. Second, we analyze the dynamics of CNNR using hardware compatible indices such as the Lyapunov spectrum, permutation entropy, spatial mutual information, average firing rate, and Kullback-Leibler divergence. We show that any one of these indices alone is not enough to reveal the dynamics in CNNR. Therefore, we combined some of these indices to evaluate the dynamics further. The relationship between the prediction performance of CNNR and combined indices is identified. We finally introduce a complexity entropy causality plane to decipher the dynamics of CNNR in detail. As a result, we provide useful qualitative methods to evaluate and design the dynamics of CNNR hardware for time-series prediction tasks.

Reservoir computing is a brain-inspired machine-learning framework that has been successfully used in information processing. A state-of-the-art methodology, called time-delay reservoir (TDR), realizes the reservoir using a single nonlinear physical node with delayed self-feedback. Memory capacity of the TDR is sensitive to the time-multiplexing procedure for input signals with a random mask. The existing memory optimization methods with respect to the masks are limited to the case without state noise, and their computational cost is large. In this article, we optimize the input mask to maximize the memory performance of the TDR with a white Gaussian state noise in a computational time efficient manner within the context of the Fisher memory curve, and then a TDR based on the Mackey-Glass system is used to illustrate our proposed method. The memory-nonlinearity trade-off of the TDRs regarding the input masks is also investigated in view of spectral properties of the spatial Fisher information matrix.

A multilayer neural network can be regarded as a type of discrete-time dynamical system in the sense that layer-to-layer information propagation can be expressed as the time evolution of a particular dynamical system. In this study, we investigate the stability of information propagation in multilayer neural networks for classification problems using finite-time maximum Lyapunov exponents, and we discuss how multilayer neural networks classify inputs. The dynamical stability analysis in this study reveals the input-dependent stability of trained multilayer neural networks. Multilayer neural networks are trained such that the information propagation is highly sensitive to input data vectors near a decision boundary for classification whereas it is less sensitive to input data vectors far from the decision boundary. This implies that the decision boundary in classification problems is characterized by a set where the finite-time maximum Lyapunov exponents of the information propagation are relatively large. These results offer a new perspective on the estimation of uncertainty of classification using multilayer neural networks.

Neuromorphic engineering is a promising computing paradigm in next-generation information and communication technology. In particular, spiking neural networks are expected to reduce power consumption drastically owing to their event-driven operation. The spike-timing-dependent plasticity (STDP) rule, which learns from local spike-timing differences between spiking neurons, is a biologically plausible learning rule for spiking neural networks (SNNs). In this study, we designed and simulated an analog circuit that reproduces the multiplicative STDP rule, which is more flexible and adaptive to external signals. We also derived analytical expressions for the behavior of the proposed circuit. These results provide important insights for designing energy efficient neuromorphic devices for applications including edge computing.

Convolutional neural network (CNN) accelerators, particularly binarized CNN (BCNN) accelerators have proven to be effective for several artificial-intelligence-oriented several applications; however, their energy efficiency should be further improved for edge applications. In this paper, a design framework for an energy-efficient BCNN accelerator based on nonvolatile logic is presented. Designing BCNN accelerators using nonvolatile logic allows for the accelerators to exhibit a massively parallel and ultra-low standby power capability. Thus, a new design can be realized for accelerators that is different from that of conventional accelerators based solely on CMOS. Considering this, we discuss a concrete design considerations of nonvolatile BCNN accelerators. In fact, a systematic design flow of the nonvolatile BCNN is established by combining Vivado HLS and standard electronic design automation tools. As a typical design example, a BCNN accelerator for inferring 32 × 32 pixel MNIST dataset is designed using a 65-nm CMOS technology. By the logic-synthesis result, the proposed BCNN accelerator is estimated to consume 94.2% lower power than that of a conventional BCNN accelerator when the frame rate is 30 frames per second.

The so-called short-term voltage stability is typical of nonlinear problems arising in dynamic performance of electric power systems. This paper reports a simulation result on load margin for short-term voltage stability of the IEEE 9-bus system connected to a multi-terminal DC system based on voltage-source converter (VSC). The simulation shows that the introduction of AC voltage control in VSC increases the load margin under a step change in load on the AC system, thereby showing a clear benefit of the AC voltage control for enhancing the short-term voltage stability without harnessing the AC system.

We demonstrated high-sensitivity measurement of photo-radiation pressure by applying frequency modulation (FM) detection for the resonance phenomenon of the micro-cantilever structure. We constructed a system to detect minute displacements using the optical-lever method and achieved a noise density of approximately 1pm/√Hz from the thermal vibration measurement. The resonance frequency shift was positive, which meant repulsive forces acted on the metal surface by way of the force generated due to incidence of the laser beam on the stainless-steel (SS) cantilever. For an incident light of 1mW intensity, a vibration amplitude of 0.687 nm and a radiation pressure of 0.378µN were measured experimentally.

An Internet of Things (IoT) employing resource-restricted (e.g., battery-limited) IoT devices at high densities will require an effective multiplexing protocol that can be implemented with low energy consumption, while being able to effectively avoid collisions between the transmissions from multiple IoT devices. Asynchronous Pulse-Code Multiple Access (APCMA) has been proposed as a communication protocol based on pulse-encoded signals that allows multiple senders to simultaneously transmit their messages. Even if multiple messages on the same band overlap in time during transmission, they are disentangled at the receiver by a decoding algorithm that is based on pattern matching of pulse trains. In this paper, we implement the APCMA protocol on an FPGA and evaluate its performance in wireless communication. We also implement a decoding algorithm that matches pulses by logical operations on a shift register. Our experiments show that the APCMA protocol can realize multiplexing with low overhead and that the error rate caused by misdetection during decoding is reduced with longer pulse trains.

A simulator that predicts the behavior of a real system is useful for reinforcement learning (RL) because we can collect experiences more efficiently than through interactions with the real system. However, in the case where there is an identification error, the experiences obtained by the simulator may degrade the performance of the learned policy for the real system. Thus, we propose a two-stage practical RL algorithm using a simulator. In the first stage, we prepared multiple premised systems in the simulator and obtained approximated optimal Q-functions for these systems. In the second stage, we represent a Q-function for the real system as an approximated linear function whose basis functions are the approximated optimal Q-functions pre-trained in the simulator. The approximated linear Q-function is learned through interactions with a real system.