Tensor decomposition is now being used for data analysis, information compression, and knowledge recovery. However, the mathematical property of tensor decomposition is not yet fully clarified because it is one of singular learning machines. In this paper, we give the upper bound of its real log canonical threshold (RLCT) of the tensor decomposition by using an algebraic geometrical method and derive its Bayesian generalization error theoretically. We also give considerations about its mathematical property through numerical experiments.

The stochastic shortest path problem (SSP) is a standard model for sequential decision making under uncertain environments. This model cannot handle cases where a catastrophic event occurs in the middle of an episode, or situations where a transition to a terminal state may not occur. Therefore, we define an SSP that takes into account dead-ends and undesired terminal states. The optimal policy for this problem is a stochastic semi-Markov policy. Since it is difficult to solve this SSP directly, an approximate problem is proposed. The optimal policy for an approximation problem is expressed as a probability distribution on a set of at most three deterministic policies. Deterministic policies are derived by considering a Bayesian-Adaptive MDP (BAMDP) for three Markov decision processes (MDPs) corresponding to the objective function and constraints regarding dead-ends and undesired terminal states. The probability distribution on a set of deterministic policies can be obtained by solving a two-person zero-sum game problem between the deterministic policies and three MDPs.

Sparse optimization with uncertainty is a widely accepted methodology that allows one to obtain robust feasible solution and then conduct sparse decision-making. To alleviate the conservative solution for robust counterpart, a probabilistic problem setup for uncertain parameter is employed to assess the risk level for the candidate solutions. Therefore, a chance constrained sparse optimization problem is well-defined that can not only measure the sparse cost but also evaluate the risk of constraints violation. In this context, we are interested in making a trade-off bridge between the sparse cost and the risk level by relaxing the constraint violations. We then shift the idea from a relaxed sparse convex optimization to risk-aware sparse optimal control application.

This paper proposes a method for estimating disturbances, which is added in the control loop of a crane with a preinstalled velocity controller, from the input-output relationship of the velocity control system. The controlled plant is a mechanically-configured velocity control system using a flow control valve. A type-1 disturbance observer was designed for the controlled plant including the velocity controller. The model parameters were identified from actual crane systems. The disturbance is assumed to be the disturbance torque generated by the wind and swaying of the suspended load applied to the crane boom. However, to avoid loss of generality, the disturbance is estimated in the dimension of angular acceleration, which is almost equivalent to torque. By feeding back the estimated disturbance via the inverse system of the velocity controller, the disturbance applied inside the controlled plant could be compensated using inputs of different dimensions.

This paper proposes a method of constructing a control system with a wide range of motion and wide bandwidth by using a two-stage mechanism equipped with a one-degree-of-freedom linear actuator, which has a narrow range of motion but can be controlled in a wide bandwidth, at the end of a general-purpose manipulator. Furthermore, a haptic interface based on variable impedance control is realized as an application of the proposed method, and its performance is evaluated. Specifically, a variable impedance control system whose characteristics change in response to orthogonal displacement was constructed by taking advantage of the fact that the manipulator can move in a direction orthogonal to the direction of movement of the two-stage mechanism.

Water environmental variables have been reasonably described as subexponentially decaying processes. In this short paper, we apply a theory of superstatistics, the statistics of doubly random variables, to recently obtained experimental data of the benthic algae detachment in water flows created in a laboratorial flume. We show that the survival curve of the algae cover (i.e., the percentage of the surface of a gravel covered by benthic algae at each time) decays subexponentially in time in each experiment with different parameter values, even under the same experimental condition. We explain the obtained experimental results through the (bivariate) superstatistics.

In this paper, a method for feed-forward cancellation of disturbance effects is proposed, and its positioning performance is evaluated and verified when the acceleration signal is acquired with high sampling. Benchmark Problem proposed by the Investigating R&D Committee on Basis of Collaborative Technologies for Precision Servo Systems is used for the verification. As a result of the verification, it is confirmed that the effect of disturbance can be reduced, and the positioning performance can be improved by using high sampling for the acquisition of acceleration signals.

This paper examines how the opinions of agents with innate beliefs change in the presence of gradually increasing peer pressure. Previous works have shown that agents’ opinions generally reach consensus as the peer pressure increases, but may oscillate without achieving consensus when the peer pressure increases too quickly. As this is a phenomenon often observed in the real world, this paper aims to clarify the characteristics of networks exhibiting such a phenomenon. In particular, for periodic connected undirected graphs, consensus is achieved when the peer pressure increases slowly, but consensus is not achieved when it increases rapidly.

The Global Navigation Satellite System (GNSS) is currently used in a variety of applications, and its positioning accuracy requirements become more and more demanding. We focus on robot which repeatedly moves and stops at low speed, and we aim to estimate its successive moving baseline vectors accurately by using GNSS. In our previous works [1,2], we proposed a method to estimate accurate successive moving baseline vectors based on the difference of the single GNSS receiver observables obtained at two successive observation times (epochs). Also, we proposed a method to estimate the bias error contained in the solution of the moving baseline vector. In this paper, based on [1,2], the methods to provide the initial value of the baseline vector calculation are modified and improved. And further experiments applied by the new proposed method and additional GNSS such as Quasi-Zenith Satellite System (QZSS) have been conducted. From the results, we can see that the proposed method can estimate the moving baseline vector within 1 [cm] accuracy.

The performance of differential evolution (DE) is influenced by various factors such as algorithm parameters and mutation strategies. JADE (adaptive DE with optional external archive) is recognized as one of the most successful studies on parameter control in DE. In this study, in order to improve the search efficiency of JADE, we propose to integrate two methods we have proposed. One is the method to estimate whether the population of candidate solutions is converging or moving for particle swarm optimization. The normalized distance between the population center and the best solution is used for the estimation. When the population is converging, the search mode is adjusted to enhance the convergence. When the population is moving, the search mode is adjusted to enhance the movement. The other is the method to control DE parameters for extreme individuals in order to improve the search efficiency. The methods are modified to use the former method in JADE and to reduce the interference between the two methods. The effectiveness of JADE incorporating the proposed methods is shown by solving thirteen benchmark problems.

We consider the high density scaling limit of a stochastic model for nonlinear diffusion which is constructed as a random walk on the discrete torus with a nonlinear diffusion rates. In this paper, we show that the deterministic limit of the stochastic model is governed by the nonlinear diffusion equation by showing the relative compactness and using the continuous mapping theorem.

Learning speaking skill is essential in language learning. It is challenging for Japanese native speaker to learn the pronunciation of vowels that are not commonly used in their native language. When pronouncing, human use their facial muscles in various way, and the usage of their facial muscles affect their pronunciation. This article contains the evaluation of Pronunciation Training System, which teach trainees by showing the formants of voice, lip shape and tongue position. Since using facial muscle, it is crucial in pronouncing vowels, in the modified pronunciation training system, indicators are added to the original pronunciation training system’s displaying of the lip shape to make trainees aware of their using of facial muscle. Evaluation was done by training experiment with 3 groups. First group uses the modified pronunciation training system, second group uses the original pronunciation training system, and last group uses no pronunciation training system. For the result, the group with modified pronunciation training system appears to use their facial muscle most effectively in articulating vowels, and the formants are more stably getting closer to the target vowel than two other groups.

The activity level of the intestine can be evaluated by the frequency of intestinal sounds generated by peristalsis, which occurs when the intestine moves and agitates its contents. Therefore, if the mechanism of intestinal sound generation frequency can be analyzed and the future state of the intestines can be predicted, various applications can be expected in the medical field, such as daily observation of intestinal health and determination of the optimal timing of medication administration. In this study, we focused on the learning interval of modeling and attempted to improve the accuracy by dividing the frequency of intestinal sound occurrence according to the state of the intestines and performing modeling.

The human intestine produces a characteristic sound called gut sound during digestion. In recent years, research has been conducted on automated bowel sound measurement systems to enable long term measurement of bowel sounds. In conventional methods, automatic bowel sound measurement has been studied using a measuring instrument equip- ed with a small microphone. However, this method involves a lot of noise from surroundings and can only be used under limited circumstances, such as for bedridden patients in intensive care units. Therefore, we decided to consider a new measurement device using a piezo electric element. In this study, we compared the performance of piezoelectric elements and diaphragms, and evaluated the ability to distinguish between intestinal sounds and noise using machine learning. deteriorates.

This study aims to enhance the reproduction of human balance movements in our previously proposed state feedback control model. One distinctive feature of human balance movements is the convex shape around the mean of the probability density function (PDF). While the previous model achieved high accuracy, it fell short in replicating these convex shapes. To address this limitation, we introduced a deadband function for control input in the new state feedback control model. Measuring balance movements in four human subjects, we identified the model parameters of the proposed model and achieved over 99% model qualities. The experimental results demonstrate that the proposed model surpassed the previous one, excelling in reproducing the convex shape observed in human’s PDFs.

Hands-free walking support tools for visually impaired people have been needed without interfering with other senses. Our study focuses on phosphene, which enables hands-free support for visually impaired people using visual information. Previous studies have reported that changing the position of electrodes placed on the face can change the presentation position of phosphenes. However, there is a limitation of the facial surface area. It is necessary to devise a combination of electrode arrangements to increase the resolution of the phosphene presentation. Therefore, this paper discusses how much the current is blocked and how the phosphene is presented at the intended location when the electrode placement across the nose is used.