In recent years, a number of studies of neural networks have been conducted with the purpose of applying engineering to the brain. Previously, we proposed a pulse-type hardware chaotic neuron model with low capacity. However, the oscillation amplitude of proposed model became smaller.
In this paper, we propose the voltage control chaotic neuron model. As a result, it is shown that the proposed model is able to achieve chaotic oscillation with sufficient amplitude. Furthermore, by constructing a neural network using the proposed chaotic neuron model, it is shown that patterns learned by STDP can be recalled.
The transmission of waveforms through optical fiber usually results in some waveform distortion. To smoothen this distortion, digital finite-impulse response (FIR) filters are commonly employed. Digital FIR filters require a relatively large area for digital signal processing (DSP) on IC chip, so an analog FIR filter is adopted. According to past researches, it is possible to realize a small area circuit by using a CMOS inverter as a delay circuit. However, number of delay circuits employing CMOS inverter increases N/2 for Multi-Level optical transmission. Therefore, we propose a new multiplier and attempted to reduce the circuit area of 4 PAM analog FIR filter. According to the estimation using SPICE simulation, it can be expected to reduce the area by 25% by using the proposed configuration.
Hypertension is one of the leading risk factors for cerebrovascular, cardiovascular, and chronic kidney diseases. Regular measurement and monitoring of blood pressure is important to decrease or prevent pathogenesis of diseases. Contactless measurement of blood pressure using a smartphone application, can enable regular monitoring. The objective of this study is to construct a system which can monitor blood pressure anytime and anywhere. In this study, contactless blood pressure assessment was attempted, using facial visible image analysis and created individual models for blood pressure estimation. Variation in brightness of the skin color was obtained from facial visible images and applied to independent component analysis, which is one of the blind source separation methods to extract the facial photoplethysmogram (PPG) component using a proposed system. Amplitude and phase of facial PPG component were used as indices for blood pressure. A correlation analysis between facial PPG component and blood pressure was performed and created individual models for blood pressure estimation.
Time-Resolved Near-Infrared Spectroscopy (TR-NIRS) is a method of analyzing the time response of the reflected light using a light diffusion equation, and enables us to measure the hemoglobin (Hb) absolute concentration, In this research, we evaluated the applicability of TR-NIRS for the measurement of brain activity. First, we conducted the simultaneous measurement of Continuous Wave Near-Infrared spectroscopy (CW-NIRS) and TR-NIRS in the working memory tasks. The results show the conventional analysis is not applicable, because the fluctuation is larger than the task-related change. Second, we proposed a new signal processing method for suppressing the fluctuation and examined the validity in the original fantom. Finally, by using the proposed method, we conducted the simultaneous measurement of TR-NIRS and functional MRI (fMRI) in the working memory tasks, and verify depth of reach and light path of near infrared light. The result suggests that we can estimate the Hb concentration due to brain activity using the proposed method.
The magnetic resonance imaging (MRI) is usually installed in a magnetically shielded room (MSR), in order to reduce the leakage flux density to less than 0.5 mT outside the room. The small space enclosed by the wall causes patients a great deal of stress and discomfort because they are isolated from the operators and from the scenery outside. In order to improve amenity for patients in hospitals, open type of MSRs for MRI, which use canceling coils, combined with square cylinders made by magnetic and conductive material, and magnetic material plates instead of magnetic walls. The magnetic shielding effects of open type of magnetically shielded room combined with square cylinders model made by magnetic, conductive material, magnetic material plates model and lattice shape model made of magnetic material for MRI are estimated by 3-D magnetic field analysis. In this report, we clarified simple lattice of magnetic plate with gap to improve shielding performance using 3D magnetic field analysis.
Selfies may have an effect on the psychophysiological state. The objective of the present study was to evaluate the psychophysiological state by providing feedback for selfies. The effect of the time period was also investigated. The period from the date a selfie was taken to the experiment date was configured as 2 conditions: a Short period (1-40 days) and a Long period (41-80 days). In this study, the evocation of emotions was attempted by providing feedback for four types of selfie 1: good face, 2: bad face, 3: good psychological state, and 4: bad psychological state. “Comfort emotion”, “Like emotion”, and “Awareness” significantly increased when providing feedback for a selfie evaluated as having a good face and a good psychological state. The experimental results using two-way analysis of variance revealed that the physiological state was not affected by the time period, but the psychological state was affected by the time period. The combination of (Positive Face Evaluation) × (Long period) selfies had more positive effect on “Awareness”.
An imbalance of the circadian rhythm due to an irregular lifestyle leads to autonomic dystonia. The long-hour variations of the physiological indices such as core temperature and heart rate variability, associated with the autonomic nervous activity, have been focused. Additionally, the relationships between these physiological indices and facial skin temperature have been reported. The objective of this study is to identify the day-long variable components of facial thermal images (FTIs) to evaluate the corresponding variations of autonomic nervous activity by non-contact measurement. In this study, the measured FTIs were subjected to independent component analysis (ICA). Additionally, multiple regression analysis was performed to estimate the relationships between independent components, extracted from FTIs, and other psychophysiological indices, associated with autonomic nervous activity. As a result, the facial day-long variable components, representing the reproducible rhythm through multiple days, were identified by applying ICA to several different combinations of FTIs. Moreover, these components were associated with the axillary temperature. Therefore, the long-period variations, associated with autonomic nervous activity, could be evaluated from the facial skin temperature.
Electroencephalogram (EEG) is generated by collective activity of neural population. In those waves, low frequency waves, called alpha waves (8-12[Hz]), are often observed in visual cortex, and the relationship between such waves and cognitive functions have been studied by many researchers. However, the mechanism of the generation of alpha waves is not understood well. In this study, we build an experimental system that can present quantitative light stimuli, and record the response of alpha waves. As a result, we observed that the onset of the recovery of alpha waves power is delayed comparing to the end of the stimuli, and also confirmed that there is a difference between decay speed and recovery speed of the alpha waves. By comparing such results with mathematical models of alpha waves already proposed, we can understand which model conforms to real alpha waves well, and that will be helpful for understanding how alpha waves are generated, and how human visual cortex works.
In this paper, we propose a method to estimate the tongue motion direction and silent speech based on convolutional neural network (CNN) using the surface electromyogram (EMG) from the suprahyoid muscles. Conventional human machine interface (HMI) is difficult to use for users who are unable to freely move the muscles below the neck due to nerve damage or the like. Therefore, we have developed a method to estimate the tongue motion in 6 directions and 5 vowels of silent speeches from 4 channel EMG. As a result of verification experiment, we obtained averaged accuracy was about 81.2% in the estimation of the tongue directions and the silent speeches. Thus, it was suggested that simultaneous estimation is possible based on EMG measured from electrodes on the anterior neck region.
Incentive Stackelberg strategy for linear parameter varying (LPV) systems with multiple decision makers is investigated. The linear quadratic control (LQC) for the LPV systems are reformulated by means of matrix inequalities. In order to decide the strategy set of multiple decision makers, Pareto optimal strategy is applied for each player. The solvability conditions of the problem are established from linear matrix inequalities (LMIs).
This paper gives mathematical expressions of three control specifications of an extended generalized minimum variance control(GMVC). The specifications are (1) the closed-loop poles including cancelled poles, (2) controller poles and (3) the steady state gain when output feedback is cut caused by a fault. The most important issue in operating plants is safety. One effective scheme for safety control is fault tolerant control which ensures plant safety when fault occurs in both of transient state and steady state. That is, in fault, (1) transient response should not be dangerous, that is, have not a large overshoot and not be strongly oscillatory, and (2) steady state gain should not be large. Transient response is settled by the closed-loop poles and the controller poles. Hence to design fault tolerant control, this paper obtains mathematical expressions of the above three specifications. GMVC had a pioneering role to model predictive control and is applied in industry. Hence, this paper obtains mathematical expressions of the specifications of GMVC. To design the controller poles, this paper uses an extended generalized controller by coprime factorization approach. Once these expressions are obtained, then we can decide design parameters straightforwardly to ensure the control specifications. These expressions include the design parameters as mathematical symbols, hence to derive the expressions, this paper uses symbolic processing software and to solve the expressions symbolically.
In this paper, an object image retrieval method based on unsupervised SLIC is proposed. To represent the object in background, local texture feature as well as global edge and color features are extracted from the extracted object region to generate unified robust feature vectors. To show the effectiveness of our method, noisy object image retrieval using real and standard image databases were executed. The precision rates obtained by the cross-validation were calculated to evaluate the performance of image retrieval. High-performance object image retrieval was achieved compared with the content-based image retrieval method using the same types of combined robust features.
This paper proposes a multi-point combinatorial optimization method based on not only higher structure solution space but also diversification and intensification strategy. The higher structure solution space is interpreted as a set of basins of attraction which is a set of solutions arriving at a same local optimal solution by best-improvement local search. Numerical experiments show that a parameter of a previous method can adjust the balance between intensification and diversification. Based on the numerical experiments, the parameter schedule for a diversification and intensification strategy are proposed. The proposed method, which has the proposed parameter schedule, aims to promote diversification at the initial stage and intensification at the final stage in the search for a basin of attraction having a superior local optimal solution, i.e., the search in the higher structure. The performance of the proposed method was evaluated though numerical experiments using benchmark problems.
We present risk word suggestion for enabling an auditor to identify potential risks based on words of identified risks in an audit report. Once the auditor describes the identified risks in the report, words related to the potential risks are inferred with words in the description of the identified risks, and suggested to the auditor. For the word inference, we assume that words of potential risks and identified risks are related each other, and apply Bayesian inference to reveal the relation between the words. In our evaluation experiments with real five cases, Bayesian inference can suggest the words for identifying potential risks. Furthermore, two auditors can accurately identify potential risks by the suggested words.
This paper presents a practice of project-based learning (PBL) class for 3rd grade undergraduate students at Nippon Institute of Technology. The paper also presents text analysis based on student activity reports for improvement of PBL teaching. To classify project activities, we introduce a new indicator that indicates whether project activity is high or low. We call the indicator project matching. Project activities are classified based on the project matching and the presence or absence of clients. We analyze each activity report of the classified project activities using text mining method. As a result, from the spring semester to the fall semester, we found that the number of words in activity report with high project matching increased, but the number of words in activity report with low project matching decreased. We also found that the target of the final product differs depending on the presence or absence of clients. We discovered that students who do not have clients need regular checks.
AHP (Analytic Hierarchy Process) is an effective method for product recommendation, because each evaluation criteria are evaluated for target objective, and each alternative are evaluated for each evaluation criteria. This method recommends alternatives to the user according to the total degrees of weight, which is calculated from these two kinds of evaluation values. The determination of weight is hard work, because of many evaluations.
This paper proposes a product recommendation system with AHP according to Normalization allocation and Hough conversion in order to evaluate alternatives. Normalization allocation can calculate automatically the evaluation value of the alternatives for every criterion. Hough conversion can extract the number of straight line elements from the target image. Through the experiment, it is confirm that the proposal system can recommend alternatives as well as the conventional system.
For dental therapy, a dentist often touches a patient face in order to identify the causes of pains such as a tumor and/or cancer. Although the importance of maxillofacial palpation has been recognized, practical training with a dental patient has not been conducted for taking care of the safety of the patient in dental school. The training of therapeutic planning with an advising doctor would be effective to enhance the quality of dental therapy. Then, this study has developed a virtual training system for maxillofacial palpation which can be utilized in dental education. In this study, we first took a multi-slice CT data from a human head mannequin in order to obtain the shape data of a virtual patient model. This study adopted Meshless Method (MLM) to construct the dynamic model of the virtual patient model. In the experimental results using a haptic interface, PHANToM Omni, the parameters of MLM influenced the surface deformation of the virtual patient model. This paper finally discussed the availability of MLM for the virtual training system and clarified the current issues that should be solved to utilize the system in dental education.
In this study, we examine properties of Ezoshika meat by bioelectrical impedance analysis (BIA) using LCR meter to estimate the freshness and aging of the meat. As a result, it is revealed that BIA measurement which is applied for the first time to Ezoshika meat is possible to evaluate the freshness and ripening degree of the meat. Also, the obtained data agreed well with the simulation results performed assuming Cole's equivalent circuit. Since this measurement can be performed quickly, it became clear that it is suitable as a simple evaluation of the meat.
This study considers the timing of human walking assistance. In the proposed method, the measurement data of knee angle is converted into the phase of oscillation model. The proper assist timing is determined in accordance with the phase information.
In motor vehicle industries, motor vehicle simulators have played important roles to develop products safely and quickly. A driver model in a simulator is required to make the vehicle speed track the reference speed in order to evaluate adequacy of lots of tuned parameters in the vehicle. The above demand has not been satisfied completely because conventional driver model outputs, opening ratio of an accelerator pedal and a break pedal stroke, have been determined based on if-then rules. In this paper, a database-driven PII2D control is applied as a driver model, and effectiveness of the method is evaluated by an actual motor vehicle simulator.
We propose traffic engineering which exists together with metric base QoS routing. In shortest paths routing, particular several links tends to be congested. TE in our proposal is adopted only to the specified area, TE area, consisting of congested links. So route of traffic which does not transmit through TE area have no change. We realize this TE routing with metric base, which means all routes are determined as shortest paths according to link costs. We formulate this as LP network problem (primary problem) and dual problem for derivation of link costs. Simulation results confirm that proposed method achieves metric base QoS routing without congestion.
A non-contact heartbeat monitoring sensor using stepped-FM UWB scheme is suggested which is robust to body movement. The biological signal can be obtained from periodic chest movement consisting of the heartbeat and respiration movement. However, the heartbeat sensing suffers from body movement because of its smaller displacement relative to breathing motion. Therefore, a stationary subject is assumed for conventional heartbeat monitoring schemes. However, some displacement of body can occur during the measurement. This paper suggests a heartbeat estimation scheme with high accuracy for some body movement. The estimation performance has been experimentally evaluated for four subjects sitting on the stool using our fabricated sensor. It is found that our proposed system can achieved the estimation error less than 2%.
Sleep apnea syndrome (SAS: Sleep Apnea Syndrome) is well-known as a disease that falls into intermittent apnea when sleeping. The estimated number of patients with SAS is known to be about 2% (2-3 million) of the Japanese population, many patients are not aware because of the disease occurring during sleep. This paper proposes a method of extracting breathing frequency necessary for SAS diagnosis using time series data of distance information on 2D image obtained from 3D vision sensor. At this time, in order to accurately estimate the breathing frequency, the position of the subject is detected from the distance information, and the frequency analysis of the distance information is performed only for the position of the detected subject. By verification using healthy subjects, the authors confirmed that the proposed method can correctly acquire the position of the subject and can estimate the breathing frequency of the subject.
IIR adaptive notch filter (ANF) with allpass filter has many advantages such as reduction in filter order and realization of flat in-band characteristics compared to ANF using adaptive transversal filter. However, the removal performance of ANF deteriorates due to the influence of the colored noise included in the input signal. In this paper we propose a cascaded IIR ANF that is not affected by colored noise. The cost function of this ANF is the square sum of correlation function of the input and output signal with the decorrelation parameter not the square sum of conventional output error. We proved that the adaptive weights of the second order IIR ANF converge near the optimum value even if the number of sine waves of the input was unknown. The numerical verifications of the convergence performance compared to the conventional LMS-IIR ANF are also presented.
Q-learning methods evaluate and update action values using information on rewards obtained. Since the Q value can not be updated until the learning succeeds and the reward is obtained, there is no index for learning, which causes a problem of requiring much time for learning. In cases, the route with no spread in the maze where the probability that learning fails is high is the semi shortest route from the start to the goal, the semi shortest route can not be learned.
To learn the optimal actions and discover the semi shortest path, it is essential to experience a large number of unknown states at early stages of the learning process. To this end, in this work we propose unknown-adventure Q-learning, in which agents maintain an action history and adventurously seek out unknown states that have not yet been recorded in this history. When unknown states are present, the agent proceeds boldly and adventurously to search these states without fear of failure. Our unknown-adventure Q-learning experiences large numbers of states at early stages of the learning process, ensuring that actions may be selected in a way that avoids previous failures.
This enables a massive acceleration of the learning process in which the number of episodes required to learn a path from start to goal is reduced 100-fold compared to the original Q-learning method. Moreover, our method is capable of discovering the semi shortest-length path through a maze even in cases where that path does not expand through the maze, a case in which learning failures are common and in which the semi shortest path cannot be discovered by methods that use V-filters or action-region valuations to accelerate learning by emphasizing prior knowledge.
We aimed at developing a simple fatigue measurement system, so far the influence of the noise under laboratory environment is small, and the experimenter side who uses a simulator. We have been studying fatigue measurement in a manageable environment. In order to show the effectiveness of our proposed method in a real environment, we collected fatigue measurement indexes before and after work for technical engineers of Software Development Company operating VDT work, Changes before and after were compared and examined. As a result, in the conventional indexes, a significant difference was found in the pulse rate and the flicker value, we found that there are significant differences between before and after some speech of fundamental frequency, speech power, period of speech indexes. In VDT work, there was a difference in flicker value as a fatigue measurement index, which was also supported by subjective symptoms and self-evaluation results. In addition, since significant differences were found also in the speech index, in the work centering on the VDT work as well as the bicycle simulator so far, it was shown that speech indexes can be effective indicators.
This paper proposes an image super-resolution technique with convolutional neural networks using horizontal and vertical filters. In the proposed method, calculation costs become small because square filters at a hidden layer are replaced with horizontal and vertical bar filters. Experimental results have shown that the average processing time for the proposed architecture was only a half of the conventional one while keeping high image qualities.