In this article, we propose the construction of a Convolutional Neural Network (CNN) model as a novel approach to detect fake images generated by diffusion models. However, constructing a CNN model to efficiently detect high-quality fake images may consume substantial computational resources, emphasizing the need for the development of a computationally inexpensive, lightweight model.At the crux of our research is the inquiry into whether a CNN can accurately detect counterfeit images produced by diffusion models using only partial image information. Further, to elucidate how this detection mechanism specifically functions, we delve deeply into the attention mechanism of the CNN.Experimental results indicate that the CNN places particular emphasis on edge information in its image analysis. This could potentially be key to effectively pinpointing specific features of counterfeit images generated by diffusion models.

Convolutional Neural Networks (CNNs) excel at various image-related tasks. These networks extract local features from images using small-sized kernels. By stacking multiple convolutional layers, they can not only focus on local regions but also capture broader global features of the input image. In this article, we’ll examine the kernels learned by various CNNs to understand the features they extract. We’ll also assess the significance of these kernels and discuss how they relate to the CNN’s performance.

In the realm of reservoir computing, a well-designed internal structure of the reservoir layer can bestow enhanced memory capacity and greater expressive prowess. However, a noteworthy trade-off exists between memory capacity and expressiveness. Consequently, even with optimal internal configurations, achieving simultaneous improvements in both aspects remains a challenging endeavor. In this study, we empirically investigate the potential performance enhancements achievable through hysteresis reservoir computing. Specifically, this approach incorporates hysteresis neurons within the reservoir layer of traditional reservoir computing.

In supervised learning, the amount of work required to assign labels to data becomes enormous as the input data increases. To reduce this work, we propose a method for label estimation using only a small amount of labeled data. The SVM separation hyperplane is moved by iterating label assignment and classification, and the results are evaluated using a new cluster evaluation criterion. This criterion is based on the Fisher criterion used for linearly separable data and applied to nonlinear data. Some semi-supervised learning methods, which are related to the proposed method, are also explained. We show that the proposed method is effective for linearly inseparable data through experiments. In addition, we compared the performance of the proposed method with some semi-supervised learning methods. The contributions of this study are the classification of data in situations where labeled data is scarce and the development of new criteria for evaluating non-linear classification results.

We proposed Nonlinear Map Optimization (NMO) to improve the solution search capability of particle swarm optimization (PSO) algorithm. NMO is one of PSO-based swarm intelligence algorithms. We have previously proposed a canonical deterministic system PSO (CD-PSO) that is removed probabilistic factors from PSO to analyze its solution search behavior and is extracted only the essential dynamics of the solution search property of PSO. Although the dynamics of CD-PSO describe the basic dynamics of PSO, the solution search performance of CD-PSO is very poor than PSO. One of the causes is the distribution of search points. Although the search point distribution of PSO has a normal distribution shape depending on stochastic factors, the search point distribution of CD-PSO which is a deterministic system is not similar to the normal distribution. In order to improve the search point distribution of CD-PSO, we proposed a modified CD-PSO. Based on the modified CD-PSO, we proposed an NMO algorithm whose search points are derived by a nonlinear map. In this article, we clarify that the solution search property of NMO. The distribution of search points is similar to a normal distribution shape. Also, the nonlinear mapping of NMO derives complex behavior due to nonlinearity depended on parameters. These properties lead to the local search capability of NMO which is improved compared to PSO. Also, information exchange within the swarm similar to PSO is related to the global search ability. Since NMO can consider local search and global search separately, the solution search capability can be improved with unimodal functions than conventional PSO.

We propose a novel optimization algorithm which named Nonlinear Map-model Optimization (abbr. NMO) method. The NMO is classified as swarm intelligence (abbr. SI) optimizer and consists of some search individuals whose dynamics is driven by a simple nonlinear map. The search point distribution is controlled by the simple nonlinear map. Based on the theoretical analysis results about the dynamics of the particle swarm optimization, we set so that the searching point distribution of the NMO becomes an optimal distribution. Also, the simple nonlinear map generates a chaotic search point time series while keeping the search range. Such a time series can efficiently search within the search range. As a result, NMO can search along the valley of the evaluation function. Namely, NMO is considered to have a rotation invariance and a scaling invariance. In general, the computation amount of SI optimizer is proportional to the number of search elements included in the SI optimizer. However, the NMO requires only a few particles comparing with other swarm intelligence optimizers. Therefore, the computation amount is the smaller than the other methods. As the result, the search performance of the NMO exhibits better than Standard PSO 2011.

Particle swarm optimization (PSO) is a stochastic population-based algorithm that is designed for real-parameter optimization problems. PSO is simple and powerful algorithm, and is applied to many real world problems. However, because the bias of the search area exists in the conventional PSO, the search performance is deteriorated in non-separable problems. In order to overcome this problem, standard particle swarm optimization 2011 (SPSO2011) was proposed. The performance of SPSO2011 is not affected by the dependencies among variables. In this article, we clarify that SPSO2011 performance is affected by the distribution of the center of the search range. Also, we clarify that the global search ability fades away by the update rule of the center. Therefore, we propose a novel update rule to improve the global search ability. We clarify the effectiveness of the proposed method by numerical experiments by using CEC2005 benchmark functions.

Particle swarm optimization (PSO) is a stochastic population-based algorithm that is designed for real-parameter optimization problems. PSO is a simple and powerful algorithm. However, the performance of PSO is degraded in the case of non-separable and ill-conditioned problems. In this article, we discuss the relation between the Hessian matrix of a function and the covariance matrix of the search distribution. The covariance matrix adaptation mechanism is required to solve non-separable and ill-conditioned problems. Therefore, in order to solve such problems, we propose a simple covariance matrix adaptation mechanism that uses the difference vector of the personal best positions. In addition, we propose a selection rule to improve the local search ability. Finally, we clarify the effectiveness of the proposed method in solving non-separable and ill-conditioned problems by using test functions.

The particle swarm optimization (abbr. PSO) is classified into one of meta-heuristics, it mimics the behavior of swarm intelligence of school of fish and flock of birds. Since the PSO is a simple algorithm, the implementation is easy. Also, the PSO does not require the gradient of the objective function. Therefore, the PSO can apply to various optimization applications. The PSO has two important control parameters: an inertia coefficient and an acceleration parameter. Especially, the inertia coefficient controls the convergence property. In order to improve the performance of the solution search ability, various kinds of control methods for the inertia coefficient are proposed. In this article, we propose a novel PSO that Lévy flight is applied to the inertia coefficient. The novel PSO is named Lévy flight PSO (abbr. Lévy-PSO). In order to confirm the performance of Lévy-PSO, we carry out some numerical simulations by using well-known benchmark function. The numerical simulation results indicate that the heavy-tailed of Lévy distribution is important to improve the search performance of Lévy-PSO.

Firefly algorithm is one of meta-heuristic algorithms. The fundamental principle of the firefly algorithm is based on the mimic of the characteristic of the blinking of natural fireflies. In this paper, we analyze the dynamics of the firefly algorithm. In order to analyze the dynamics rigorously, we apply a deterministic system that is removed the stochastic factors from the state update equation. The behavior of the individual deterministic firefly exhibits a chaotic behavior depending on the parameters. Therefore, we investigate relationship between the behavior of the firefly and the parameters. Also, we investigate the effect that the chaotic behavior influences the searching ability.

In order to analyze the dynamics of the particle of the particle swarm optimization (abbr. PSO) rigorously, we proposed a canonical deterministic PSO (abbr. CD-PSO). The CD-PSO can be described a quite simple equation, and it is very easy to analyze the dynamics. However, the CD-PSO is a deterministic system, therefore, the solution search ability is worse than the conventional PSO which contains stochastic factors. The deterministic system is easy to implement since stochastic factors are not contained. Therefore, we consider the improvement method of the search ability for the CD-PSO. Based on the analysis results of the CD-PSO, we propose a deterministic multi-agent solving method (abbr. MAS). To improve the solution search performance, we propose a novel asynchronous MAS whose update manner is asynchronous. By using some benchmark functions, we confirm the effectiveness of the asynchronous MAS.

Particle swarm optimization is very simple algorithm. However, the trajectory of each particle becomes pretty complicated because each particle is affected by the information of the other particles. In order to analyze the dynamics of the PSO rigorously, we propose the canonical deterministic PSO (CD-PSO). The CD-PSO is described by a canonical form, and the dynamics is characterized by the damping factor and the rotational angle. The dynamics of the CD-PSO can be analyzed theoretically, however, the solution search performance is worse than the conventional PSO. The reason why the performance of the CD-PSO is poor, is the rotation radius of the particle is decreased with the change of the best position. The small rotation radius contributes to the local search ability, but the global search ability is spoiled. Therefore, we propose a method to sustain the global search for the CD-PSO. The function is realized by maintaining the rotation radius. Applying the proposed method for the CD-PSO, the solution search performance improves that of the conventional CD-PSO.

This article analyzes the convergence property of the particle swarm optimization and its application to the nonlinear blind source separation system. The inter-particle communication of the particle swarm optimization is realized by the past history of the neighbors and depends on the network structure of the swarm. We focus on an average path length of the network, and we clarify the relationship between the average path length and its searching performance. The result indicates that a long average path length is effective for multi-modal functions and multi-optima problems. Therefore, we apply the PSO with the long average path length to a nonlinear blind source separation system. Blind source separation is a technique for recovering an original source signal from mixing signals without the aid of information of the source signal. The system restores the original signal using the probability of the distribution of the original signal. In this study, we consider the case where the original signals are nonlinearly mixed. In general, the separation of the nonlinear mixture signals is quite difficult. In order to solve such problem, we apply a radial basis function network to the nonlinear blind source separation system. The radial basis function network can approximate the nonlinear mapping. Therefore, the inverse mapping of nonlinear mixture system is approximated by the RBF network. For the system to be able to approximate the inverse mapping, it is necessary to learn the parameter of the RBF network. PSO is used for a learning algorithm. Simulation results show that the proposed approach has good performance.

This paper discusses a particle swarm optimization (PSO) with switched topology and its application to the multi-solution problems. First, we introduce a deterministic PSO characterized by normalized deterministic parameters and a canonical form system equation. This system is convenient to grasp effects of parameters on the stability. Second, we investigate effects of the average distance of several the swarm topologies on the search capability. Especially, we introduce the switched topology where any information is not transmitted from the edge if the switch is off. Third, we consider an application to exploring multiple periodic points in simple dynamical systems. Performing numerical experiments for typical examples, the algorithm performance is investigated.

Particle swarm optimization (abbr. PSO) is one of the most effective optimization algorithms. The PSO contains many control parameters, therefore, the performance of the searching ability of the PSO is significantly alternated. In order to analyze the dynamics of such PSO system rigorously, we have analyzed a deterministic PSO (abbr. D-PSO) systems which does not contain any stochastic factors, and its coordinate of the phase space is normalized. The found global best information influences the dynamics. This situation can be regarded as the full-connection state. On the other hand, there is the case where the best information in a limited population. Such information is called as lbest. How to get the lbest information from any population is equivalent to a network structure. Such network structure influences the performance of searching ability. In order to clarify a relationship between network structures of the PSO and its performance, we pay attention to the degree and the average distance used in graph theory. We consider the two cases where the D-PSO has an extended cycle structure and a Small World network structure. Our numerical simulation results indicates the searching performance of the D-PSO is depended on the average distance of the node. Especially, the long average distance exerts the search performance on the D-PSO. We confirm that the search performance properties of the D-PSO and the conventional stochastic PSO are completely different to the average distance. The search performance of the D-PSO is improved according to the average distance. On the other hand, the search performance of the conventional stochastic PSO is deteriorated according to the average distance. We consider that the slow transmission of the beneficial information leads to the diversification of the particles of the D-PSO. Also, we clarify the small perturbation of the random range of the stochastic PSO is important.

This paper studies the particle swarm optimization (PSO) with switched topology and its application to the multi-solution problems (MSPs). The switched topology can give variety in the swarm structure and can be effective to control information transmission speed. We introduce the switched distance as a basic measure to evaluate the switched topology. The proposed PSO does not include stochastic parameters and is suitable for theoretical analysis. Applying the proposed PSO to typical benchmarks of the MSPs, the algorithm efficiency is investigated.

This article discusses a design procedure of DC AC inverter. The DC-AC inverter is to produce a sinusoidal AC voltage with adjustable amplitude and frequency. Pulse-width modulation (abbr. PWM) is one of the most used techniques in static inverters. For the PWM, the switching angle is most important, and the switching angle controls the efficiency of DC AC inversion. For this reason, the design of the optimal switching angle vector is very important. In this article, we obtain such switching angle vector by particle swarm optimization system (abbr. PSO). Our simulation results indicate that the proposed design procedure gives high efficiency inversion.

Particle swarm optimization (abbr. PSO) is one of the most effective optimization algorithms. The PSO contains many control parameters. These causes, the performance of the searching ability of the PSO is significantly alternated. In order to analyze the dynamics of such PSO system rigorously, we proposed a canonical deterministic PSO (abbr. CD-PSO) systems which does not contain any stochastic factors, and its coordinate of the phase space is normalized. The found global best information influences the dynamics. This situation can be regarded as the full-connection state. On the other hand, there is the case where the best information in a limited population. Such information is called as lbest. How to get the lbest information from any population is equivalent to a network structure. Such network structure influences the performance of searching ability. In order to clarify a relationship between network structures of CD-PSO and its performance, we pay attention to the degree and the average distance used in graph theory. First, we consider the case where the CD-PSO has an extended cycle structure. Our numerical simulation results indicates the searching performance is depended on the average distance of the node, and the optimal average distance is existed. Next, we consider the case where the CD-PSO has a Small World network structure. The extended cycle structure has uniform symmetric property. On the contrary, a small world network has nonuniform property. Even in the case where the CD-PSO has the small world network structure, the searching performance is depended on the average distance

This paper proposes a scheme to solve issue of unfair throughput by the capture effect. Wireless LAN (WLAN) systems suffer from the hidden terminal problem, which is occurred by the unreached signal for non-neighbor radio terminals. Since the CSMA/CA (Carrier Sense Multiple Access with Collision Avoidance) for WLANs does not adequately solve this problem, transmitted packets collide more frequently than is desirable, which degrades the throughput performance. Moreover,the differences in distance between terminals can trigger the near-far problem. The near-far problem emphasizes the reception of the strongest signal over weaker signals due to the capture effect. In environments that exhibit both the hidden terminal problem and the near-far problem, there arises the possibility of unfair transmission among the terminals. We propose a scheme that ensures fairness in terms of throughput.

A particle swarm optimization (PSO) system is one of the powerful systems for solving global optimization problems. The PSO algorithm can search an optimal value of a given evaluation function quickly compared with other proposed meta-heuristics algorithms. The conventional PSO system contains some random factors, therefore, the dynamics of the system can be regarded as stochastic dynamics. In order to analyze the dynamics rigorously, some papers pay attention to deterministic PSO systems which does not contain any stochastic factors. According to these results, the eigenvalues of the system impinge on the dynamics of the particles. Depending on the parameter, the searching ability of the deterministic PSO is decreased. Also, the eigenvalue is complex conjugate number, the system exhibits remarkable searching ability. In order to overcome this, we propose a canonical deterministic PSO which can control its eigenvalues easily, and can improve the searching ability. The dynamics of the system can characterize the damping factor and the rotation angle which can derive from its eigenvalue. We will confirm relation between these parameters and the searching ability of the optimal value from some numerical simulations.

We propose a multi-hysteresis voltage controlled current source (multi-hysteresis VCCS). The multi-hysteresis VCCS consists of multiple single-hysteresis VCCSs in parallel. The multi-hysteresis VCCS can exhibit various kinds of i - v characteristics. In addition, we introduce a chaotic oscillator by applying the multi-hysteresis VCCSs. The proposed oscillator is suitable for the IC implementation. A fully-differential multiscroll chaotic oscillator circuit is designed. The SPICE simulation results confirm the multi-hysteresis characteristics and various chaotic attractors.

We propose hysteresis neural networks for solving NP-hard problems, Traveling Salesperson Problems (TSP). Since hysteresis neural networks have no local minimum, they can be applied to various optimization problems. However, since conventional system for solving TSP which is proposed by Hopfield and Tank (1986) is not adaptive for hysteresis neural networks, they have not yet been applied to TSP. So, we propose a novel system for solving TSP. We obtain an improved result for the TSP with the proposed system based on hysteresis neural networks.

Many researchers have proposed combinatorial optimization problem solver by using neural networks. In this paper, we propose a synthesis procedure for hysteresis neural networks whose equilibrium points correspond to an optimal solution of the combinatorial optimization problem. This system does not constrain its energy from decreasing monotonously, namely the output of this system may oscillate. However, our synthesis procedure guarantees that all equilibrium points correspond to an optimal solution of the combinatorial optimization problem. We control the time constant of each hysteresis neuron, and restrain the system from oscillating.

In order to analyze artificial neural networks which can treat dynamical information, we consider the simple hysteresis network whose cross connections are uniform. Also, it has only two parameters: self feedback and DC term. For this simple hysteresis network, we analyze discontinuous period doubling like bifurcation set. Even if all self feedback parameters are same value, we discover chaotic response which is confirmed by Lyapunov exponents and continuous spectrum. In our previous works, we have analyzed much simpler cases. In this paper, we analyze bifurcation phenomena by numerical experiments and laboratory measurements.

This paper discusses occasional linear connection (ab. OLC) that realizes in-phase, lagging phase and leading phase master-slave synchronization of chaos. Also, occasional proportional feedback (ab. OPF) is a basic version of OLC and can stabilize desired UPO in chaos. The OLC and OPF can realize various spatiotemporal dynamics. An implementation example of the OLC is also shown.

In order to approach the explication of rich dynamics from a large scale nonlinear network, this paper analyses a continuous time hysteresis network whose cross connection is uniform. This system has only two parameters. Applying an efficient piecewise linear technique for a binary hysteresis element, we can clarify the existence condition of periodic orbits and the number of them. Also we can prove a sufficient condition for local convergence. In addition, the global convergence is verified by both numerical and laboratory experiments.>