The Hamiltonian in the fundamental equations used to model online user dynamics consists of the sum of two matrices that generally cannot be diagonalized simultaneously. Therefore, except for exceptional cases where networks are regular graphs, the eigenvalues of the Hamiltonian cannot be derived from the individual eigenvalues of those two matrices. This paper provides a derivation of the eigenvalues and eigenvectors of the Hamiltonian by considering that of the matrix obtained by squaring the Hamiltonian.

Gaussian-Bernoulli restricted Boltzmann machine (GBRBM) is a bipartite-type Markov random field consisting of two types of layers, namely, visible and hidden layers, and it can treat continuous data points. For the implementation of GBRBM (e,g, for learning and inference), the evaluation of the expectations of variables is critical; however, this is difficult because of the problem of combinatorial explosion. In this study, we propose an effective method for expectation evaluation on a (canonicalized) GBRBM based on spatial Monte Carlo integration and marginalized-space Gibbs sampling (mGS). Here, mGS is a collapsed-Gibbs-sampling version of the layer-wise blocked Gibbs sampling method and it can reduce the relaxation times of Gibbs sampling. The results of numerical experiments demonstrate the effectiveness of the proposed method.

The spread of social media has caused the problem of opinion polarization as an unexpected side effect. Therefore, it is important to elucidate the mechanism of opinion formation through communications on social media. In this paper, we clarify the relationship between the degree of similarity among topics and opinion formation in multi-dimensional opinion formation that deals with multiple topics. For this objective, we use the multi-dimensional opinion formation model that we have proposed in the past. This model has the advantage of being able to avoid the “curse of dimensionality” and of being able to handle relationships among topics. We focus on how much the formation of an opinion on a particular topic influences the opinion held on other topics. Numerical experiments show that the higher the similarity between topics, the stronger the influence.

Estimating and controlling nonlinear neuronal system are crucial for understanding the neuronal dynamics and brain functions. However, it is challenging to control the nonlinear system including unobservable state and unknown dynamics. We propose a framework for estimating and controlling an individual neuron by leveraging the sequential Monte Carlo method (SMC). We derive an online algorithm based on the expectation-maximization algorithm and constitute the control law by employing the SMC-based model predictive control. We verify the effectiveness of the proposed method using simulation environments. The results suggest we can simultaneously estimate the latent variables and the parameters and control neuronal state.

In recent years, artificial intelligence (AI) has attracted attention for edge AI, which operates in an offline environment without using clouds and focuses on the speed of responsiveness. Devices for this purpose must be power efficient and compact, and various methods have been studied to implement AI in hardware. In this context, an approximate computation method called stochastic computing (SC) can implement multiplication and addition, which are frequently used in AI computations with low resources, and are suitable for parallel processing. However, SC-based AI hardware implementations have problems with the use of conventional memory. To solve this problem, we propose a stochastic memory (SM) model based on a bistable system and simple analog circuit. We evaluate the SM characteristics using SPICE simulations and experiments using actual ICs. This is expected to enable the implementation of a combination of SM and AI hardware using SC from previous studies that are more suitable for edge AI requirements.

Reservoir Computing (RC) is a machine learning framework inspired by nonlinear science, and expected to provide a solution for edge computing, owing to its simple algorithms. Therefore, the development of edge-implementable RC enables fast and lightweight information processing in edge applications. This report introduces two technologies for achieving resource efficiency and expanding nonlinear capacity in FPGA implementations of reservoir computing. We report the effect of the proposed technologies and implemented architecture and checked its architectural features and benchmark scores. In addition, as an application demonstration, we applied our system to the prediction of a chaotic dynamical system.

A Gaussian-Bernoulli restricted Boltzmann machine (GBRBM) is often used in semi-supervised anomaly detection (AD), in which the GBRBM is trained using only normal data points. The GBRBM-based AD is performed based on a score that is identical with an energy function of the marginalized GBRBM. However, it is difficult to set a threshold of the score for discriminating between a normal and an anomaly to an appropriate value because we do not equip a valid interpretation for the score value. To gain the interpretation, we focus on features of the score: the average, variance, and minimum values; and propose a sampling-based method for evaluating the features. Numerical experiments demonstrate that the proposed method can evaluate these three quantities with a high accuracy.

This research introduces an innovative approach to automatically analyzing of non-smooth dynamical systems involving continuous and discrete components. The core of this approach is the “mode concept,” a framework that systematically separates the original systems into primitive parts and combines them with the proper rule. The study successfully constructs diffeomorphisms representing entire systems by applying the mode concept, equivalent to Poincaré maps. It implies that we can numerically obtain the Jacobian matrix and Hessian tensor of the map and use them to analyze the system behavior. The algorithm requires only the modes and their transition rules. In numerical experiments, the algorithm is implemented in Python and applied to the typical non-smooth dynamical systems: the Izhikevich neuron model, the forced Alpazur oscillator, and even the smooth systems: the Duffing equation and the Hénon map. The rich analysis results, such as the bifurcation diagrams, return map, and eigenvalue transition, are obtained using the algorithm.

Complex phenomena are observed in various situations, and might be generated by deterministic dynamical systems or stochastic systems. Clarifying and analyzing complex phenomena is an important issue in the development of various technologies, such as control and prediction. In this study, we propose a method for quantifying the complexity of graph structures obtained from chaotic time series data based on Campanharo's method. Our results show that it is possible to quantify periodic, quasi-periodic, and chaotic states from the graph structure, and that numerical values show the same tendency as the Lyapunov exponent. Furthermore, we find that specific graph patterns are generated around the period-doubling bifurcation, which shows that our method can capture characteristic features of time series data that cannot be captured using the Lyapunov exponent.

A traffic matrix (TM) plays an essential role in many network analysis tasks. Since obtaining the full TM by direct observation is challenging, many studies have focused on recovering the TM from the partial observation. These existing methods, which achieve high recovery accuracy using the spatio-temporal characteristics of TM, are computationally expensive and/or require prior training. We propose a new TM completion method based on a nonlinear, nonconvex optimization for a weighted tensor nuclear norm minimization with tensor construction based on the intrinsic periodicity of TM. Our tensor construction method does not require complicated spatio-temporal characteristic estimation and prior training. The experimental results on two real-world traffic data and various data-missing scenarios show that the proposed method can achieve the comparable recovery capability as the conventional methods with a significantly simpler problem formulation.

Semiconductor technology has been growing rapidly. In particular, mobile devices, which include smartphones and wearable devices, are widely used in daily life. In addition, the mobile devices perform digital image processing, audio processing, and artificial intelligence (AI) processing, etc., which are known collectively as multimedia processing, and require massive data calculation. Therefore, mobile devices must have high performance, programmability, and versatility on a single core. In this paper, a content addressable memory-based massive-parallel SIMD matrix core (CAMX) is proposed as a mobile device accelerator for high performance, programmability, and versatility. The CAMX can process repeated arithmetic and table-lookup coding operations in a massively parallel form. In addition, the authenticity of digital images is becoming more important. Hence, watermarking technology using a max-plus algebra-based morphological wavelet transform (MMT) is used as an authentication technique for mobile devices. In this paper, the CAMX is implemented and the MMT watermarking technology is simulated through a lookup-table. The results show that the CAMX can process 128 × 128 pixel images in parallel, where decomposition is 2,364 clock cycles and reconstruction is 1,236 clock cycles. In addition, the processing time for the CAMX and an ARM core, which is commonly used in mobile devices, are compared. The CAMX achieves higher MMT watermarking performance at operating frequencies of 800 kHz or higher as compared to the ARM core with NEON.

This study focuses on relational data obtained through object relations. Traditional analysis of relational data often ignores attribute information. Therefore, Mikawa et al. proposed a method to estimate the latent structure of continuous relational data using a generative model and parameter estimation. However, real-world relational data can be discrete, and therefore, we propose a new model for binary relational data using a generative model based on the Bernoulli distribution and the Monte Carlo Expectation-Maximization (EM) algorithm for parameter estimation. We also clarify the effectiveness of the proposed model through simulation experiments using artificial data and real data.

This paper presents a new architecture to give personality to robots and artificial intelligence (AI). Communication robots are used for elderly care and as AI assistants. However, the currently available communication robots can only perform general actions and have limitations in following instructions. In this study, we developed a new architecture that combines Ternary Content Addressable Memory with Individuality (ITCAM) and reservoir computing, which reflects chip variations in search results. This architecture is used to perform learning and inference by using the weights of nine field-programmable gate array chips of the same standard. In experiments using sine waves and triangular waves, the individuality range of each chip performed within a 5% range. This architecture allows individual chips to vary their predictions while referring to teaching data. Furthermore, experiments using atmospheric temperature show the promise that an ITCAM-based Reservoir Computing Architecture (IRC) approach can be used to predict the actual conditions in the volatile real world. Even in years where the normalized average temperature changed significantly including cases with variances of more than 10% per day compared to previous years, the IRC model can generate results within a range of 4%.

We propose a traffic prediction method for UAV networks considering connection errors to access points (APs), and demonstrate its effectiveness by calculating the average throughput. In real networks, the performance of devices that serve as APs varies, which may lead to a concentration of loads on high-performance APs. In addition, UAV networks are at risk of communication disconnections and connection errors because UAVs have different altitudes compared to ground-based networks. In this paper, we focus on the possibility that the overall system throughput may change as a result of accidents such as path changes due to the influence of wind, which cause the connection destination to shift to a lower-performance AP, by distributing the load. We use the stochastic evolutionary game theory to analytically solve the changes in throughput for each connection error probability. Moreover, we analyze the characteristic of UAV network with error probability. We also considered the case where there is a difference in altitude between the UAV and the AP, and measured the change in throughput with altitude. We also consider the case where the difference in altitude between the UAV and the AP was tied to the connection error probability.

In this paper, we explore the intricate relationship between colors within the latent variable space, employing an image generation model tailored to produce images in response to textual prompts structured as “color + car type + car name”. Upon leveraging this model to analyze the interrelation of colors across various car types, it was observed that color positions within the latent variable space showcased remarkable consistency across car categories, evoking the semblance of an equilateral triangle. Moreover, for specific car types, three RGB color points were identified as emblematic and were utilized to construct a plane. By modulating the latent variables within this established plane, it was discerned that while the car's shape remained invariant, the color underwent modifications. This leads us to postulate the potential existence of hyperplanes in the latent variable space, each emblematic of distinct car classifications. This investigation seeks to elucidate the interplay between color and shape information encapsulated within the latent variable space.

One of the neuron models that simulate the electrical activity of neurons, the multi-compartment model, has spatial electrical properties that control nonlinear spatiotemporal dynamics and can reproduce nonlinear electrical responses with high accuracy. However, it is difficult to determine the model parameters in multi-compartment models from membrane potentials, since unknown high dimensional parameters for spatial electrical property should be estimated using incomplete observation data. In this paper, we propose a data-driven method to estimate the spatial electrical properties in the multi-compartment model from membrane potentials observed incompletely. The proposed method employs the replica exchange method using prior information considering morphological smoothness to solve problems of the local optima in the solution space and incompleteness of observation data. We further verify the effectiveness of the proposed method by using simulation data obtained from realistic neuron models.

Multi-objective Optimization Problems (MOPs) attempt to find solutions simultaneously by minimizing or maximizing multiple objective functions in a trade-off relationship. As a solution method for MOPs, Cooperative Multi-Objective Differential Evolution (CMODE) has been proposed. Although the CMODE shows good performance for the MOPs, this method sometimes found partially biased solutions in a single trial. To solve this problem, we have already proposed an improved CMODE method. In this study, we comprehensively evaluate the performance of the improved method on low-dimensional and high-dimensional benchmark problems using various evaluation measures. Numerical results show that the improved method exhibits a small inverted generational distance for all low-dimensional and some high-dimensional benchmark problems, as compared to conventional solution methods.

Reservoir computing-based control is drawing focus in the field of computational intelligence. Despite the efficacy of reinforcement learning for autonomous capability enhancement, the computational overhead associated with extensive trial-and-error remains a considerable drawback. To address this issue, we introduce a novel mental simulation model underpinned by reservoir computing. This model employs an internal environmental representation to facilitate direct action sequence optimization. We rigorously evaluate the proposed framework across classic control tasks as well as a specialized application scenario under three distinct conditions: fully-observable, partially-observable, and visual-observation conditions. Our findings indicate that the proposed model outperforms prior methods in action planning.

The field of physical reservoir computing, which harnesses the dynamics of physical systems for information processing, is undergoing continuous development. The use of cultured neuronal networks as a physical system within physical reservoir computing has played a crucial role in clarifying the connection between neuronal network structure and information processing. In this study, we introduce a model that combines predictive coding and reservoir computing within cultured neurons, serving as a model for sensory information processing. Our findings indicate that the interplay between neuronal network structure and the number of neurons has a significant influence on sensory information processing. This research lays the groundwork for future inquiries into information processing within biological neuronal networks.

In recent years, neural networks have garnered significant interest due to their capacity to learn from data examples pertaining to various tasks, subsequently demonstrating proficiency in addressing these challenges. The efficacy of a neural network not only hinges on the caliber and volume of the data samples but also crucially on the architecture of the network itself. To date, a concrete theoretical approach to discern the optimal structure of neural networks remains elusive. As a result, much of the architectural design relies heavily on the expertise and intuition of the designer. Furthermore, empirical validation is indispensable to ensure the network operates as anticipated. This design paradigm often culminates in extended experimental durations. In an endeavor to mitigate this challenge, we explore the feasibility of predicting image classification accuracy post-training by leveraging information gleaned from the neural network's initial state, specifically for a designated image classification task. Employing multiple regression analyses, grounded in our antecedent knowledge, we aim to project the image classification accuracy subsequent to 150 training epochs. Our empirical findings attest to the viability of this approach, elucidating that the classification accuracy at the terminal training epochs can be forecasted with minimal margin of error.

A McKibben-type pneumatic actuator (MPA) is a type of soft actuator. Several attempts have been made to control robots mechanically driven by pneumatic actuators, and several robotic movements have been realized. Air-driven pneumatic circuits are among the methods proposed for mechanical control. Previous studies confirmed that pneumatic circuits can generate various patterns of self-excited oscillations. These self-excited oscillations can be used for periodic motion of robots with MPAs. To enhance the control of robots with MPAs, it is necessary to analyze the mechanism of the behavior generated by pneumatic circuits. In this study, we analyze the behavior of self-excited oscillations in two types of pneumatic circuits from previous studies by considering them as logical circuits. Consequently, the mechanism by which these pneumatic circuits generate periodic motion was clarified, and the actual behavior was theoretically supported. In addition, we designed a new pneumatic circuit that enabled the robot to walk based on the results of the theoretical analysis as well as verified its motion.

A cloudlet is a cluster of computers that exists within the same Local Area Network (LAN) and can be accessed by all users connected to the network in a single hop. Compared to the traditional edge computing system, where servers are located at base stations, cloudlets enable real-time communication with less network latency. On the other hand, the lack of computing power is another problem to be solved. Conventional research examines offloading jobs between cloudlets in order to reduce workload variance whilst ensuring that the total latency of each cloudlet remains below an acceptable level. However, the calculation of total latency is dependent on the fraction of offloaded jobs, which results in an unfair decision regarding latency for a smaller fraction of jobs within the system. It is necessary to explore fairer solutions for offloading decisions in cloudlet systems. In this paper, we present a method for load balancing that prioritises reducing latency, utilising game theory. We assess the method's effectiveness by comparing the differences in utilisation between cloudlets.

Message-passing (MP) is a powerful tool for finding an approximate solution in optimization. We generalize it to nonlinear product-sum form, and numerically show the fast convergence for the minimum feedback vertex set and the minimum vertex cover known as NP-hard problems. From the linearity of MP in a logarithmic space, it is derived that an equilibrium solution exists in a neighborhood of random initial values. These results will give one of the reason why the convergence is very fast in collective computation based on a common mathematical background.

Formation control of multi-agent systems has been a prominent research topic, spanning both theoretical and practical domains over the past two decades. Our study delves into the leader-follower framework, addressing two critical, previously overlooked aspects. Firstly, we investigate the impact of an unknown nonlinear manifold, introducing added complexity to the formation control challenge. Secondly, we address the practical constraint of limited follower sensing range, posing difficulties in accurately localizing the leader for followers. Our core objective revolves around employing Koopman operator theory and Extended Dynamic Mode Decomposition to craft a reliable prediction algorithm for the follower robot to anticipate the leader's position effectively. Our experimentation on an elliptical paraboloid manifold, utilizing two omni-directional wheeled robots, validates the prediction algorithm's effectiveness.