Search

With the increasing importance of data privacy protection, various privacy-preserving machine learning methods have been proposed. In the clustering domain, various algorithms with a federated learning framework (i.e., federated clustering) have been actively studied and showed high clustering performance while preserving data privacy. However, most of the base clusterers (i.e., clustering algorithms) used in existing federated clustering algorithms need to specify the number of clusters in advance. These algorithms, therefore, are unable to deal with data whose distributions are unknown or continually changing. To tackle this problem, this paper proposes a privacy-preserving continual federated clustering algorithm. In the proposed algorithm, an adaptive resonance theory-based clustering algorithm capable of continual learning is used as a base clusterer. Therefore, the proposed algorithm inherits the ability of continual learning. Experimental results with synthetic and real-world datasets show that the proposed algorithm has superior clustering performance to state-of-the-art federated clustering algorithms while realizing data privacy protection and continual learning ability. The source code is available at https://github.com/Masuyama-lab/FCAC.

Accurate and reliable relative humidity forecasting is important when evaluating the impacts of climate change on humans and ecosystems. However, the complex interactions among geophysical parameters are challenging and may result in inaccurate weather forecasting. This study combines long short-term memory (LSTM) and extreme learning machines (ELM) to create a hybrid model-based forecasting technique to predict relative humidity to improve the accuracy of forecasts. Detailed experiments with univariate and multivariate problems were conducted, and the results show that LSTM-ELM and ELM-LSTM have the lowest MAE and RMSE results compared to stand-alone LSTM and ELM for the univariate problem. In addition, LSTM-ELM and ELM-LSTM result in lower computation time than stand-alone LSTM. The experiment results demonstrate that the proposed hybrid models outperform the comparative methods in relative humidity forecasting. We employed the recursive feature elimination (RFE) method and showed that dewpoint temperature, temperature, and wind speed are the factors that most affect relative humidity. A higher dewpoint temperature indicates more air moisture, equating to high relative humidity. Humidity levels also rise as the temperature rises.

Biologically inspired intelligent methods have been applied to various sensing systems in order to extract features from a huge size of raw sensing data. For example, point cloud data can be applied to human activity recognition, multi-person tracking, and suspicious person detection, but a single RGB-D camera is not enough to perform the above tasks. Therefore, this study propose a 3D environmental map-building method integrating point cloud data measured via multiple RGB-D cameras. First, a fast multi-scale of distributed batch-learning growing neural gas (Fast MS-DBL-GNG) is proposed as a topological feature extraction method in order to reduce computational costs because a single RGB-D camera may output 1 million data. Next, random sample consensus (RANSAC) is applied to integrate two sets of point cloud data using topological features. In order to show the effectiveness of the proposed method, Fast MS-DBL-GNG is applied to perform topological mapping from several point cloud data sets measured in different directions with some overlapping areas included in two images. The experimental results show that the proposed method can extract topological features enough to integrate point cloud data sets, and it runs 14 times faster than the previous GNG method with a 23% reduction in the quantization error. Finally, this paper discuss the advantage and disadvantage of the proposed method through numerical comparison with other methods, and explain future works to improve the proposed method.

Abstract We have developed a stress management framework aimed at helping healthcare professionals manage occupational stress. A chatbot is used to gather pertinent information from a user through regular conversation which is processed via a fuzzy inference model to derive their level of occupational Comprehensibility, Manageability, Meaningful and overall occupational Sense of Coherence (SOC). This analysis feeds into a Peer Support model which selects the best intervention aimed at enhancing a user’s stress-coping abilities and reducing their levels of occupational stress. A trial has been conducted with working adults, and the results were examined using a questionnaire supported by Technology Acceptance Model, which showed that the chatbot could recognize the user's SOC, but left a challenge: few users wanted to continue using it in their daily lives.

Computer vision-based cyber-physical-social systems (CPSS) are predicted to be the future of independent hand rehabilitation. However, there is a link between hand function and cognition in the elderly that this technology has not adequately supported. To investigate this issue, this paper proposes a multiscopic CPSS framework by developing hand–object interaction (HOI) based on visual attention. First, we use egocentric vision to extract features from hand posture at the microscopic level. With 94.87% testing accuracy, we use three layers of graph neural network (GNN) based on hand skeletal features to categorize 16 grasp postures. Second, we use a mesoscopic active perception ability to validate the HOI with eye tracking in the task-specific reach-to-grasp cycle. With 90.75% testing accuracy, the distance between the fingertips and the center of an object is used as input to a multi-layer gated recurrent unit based on recurrent neural network architecture. Third, we incorporate visual attention into the cognitive ability for classifying multiple objects at the macroscopic level. In two scenarios with four activities, we use GNN with three convolutional layers to categorize some objects. The outcome demonstrates that the system can successfully separate objects based on related activities. Further research and development are expected to support the CPSS application in independent rehabilitation.

Growing neural gas (GNG) has been widely used in topological mapping, clustering and unsupervised tasks. It starts from two random nodes and grows until it forms a topological network covering all data. The time required for growth depends on the total amount of data and the current network nodes. To accelerate growth, we introduce a novel distributed batch processing method to extract the rough distribution called Distributed Batch Learning Growing Neural Gas (DBL-GNG). First, instead of using a for loop in standard GNG, we adopt a batch learning approach to accelerate learning. To do this, we replace most of the standard equations with matrix calculations. Next, instead of starting with two random nodes, we start with multiple nodes in different distribution areas. Furthermore, we also propose to add multiple nodes to the network instead of adding them one by one. Finally, we introduce an edge cutting method to reduce unimportant links between nodes to obtain a better cluster network. We demonstrate DBL-GNG on multiple benchmark datasets. From the results, DBL-GNG performs faster than other GNG methods by at least 10 times. We also demonstrate the scalability of DBL-GNG by implementing a multi-scale batch learning process in it, named MS-DBL-GNG, which successfully obtains fast convergence results. In addition, we also demonstrate the dynamic data adaptation of DBL-GNG to 3D point cloud data. It is capable of processing and mapping topological nodes on point cloud objects in real time.

The intricate dynamics of the human musculoskeletal system require complex mathematical computations for accurate simulation, posing challenges in estimating muscle activity. Real-time processes and comprehensive analysis greatly influence the effectiveness of monitoring applications. The objective of our research was to enhance real-time muscle activity predictions by incorporating environmental data into human musculoskeletal simulations, focusing on the upper extremity. Our model, developed using MuJoCo software, consisted of 50 Hill-type muscles and integrated environmental context. Information on human posture was collected from single RGBD sensors positioned at 32 three-dimensional node locations. We used inverse kinematics computations to convert this data into joint angle parameters for our simulation model. The stretch reflex of each muscle was regulated to initiate movement in the target joints. Desired muscle stretch length was derived from the mechanical interaction between the bone structure and the muscle-tendon actuator connected to it. Our model also allowed for the application of artificial force to simulate external load conditions. To validate our model, we performed basic movements with the upper extremity and measured muscle activity using EMG sensors. Our results confirmed the model’s ability to accurately predict muscle activation and the force exerted by each muscle. Further experiments demonstrated its potential for seamless integration with dynamic environmental conditions, thereby enhancing its utility as a comprehensive human physical monitoring system.

This paper introduces a multiscopic cyber-physical-social system (CPSS) to bridge the gap between independent rehabilitation in physical and cognitive aspects. Specifically, we focus on hand–object interaction (HOI) recognition with visual attention for the block-design test (BDT). The proposed framework utilizes three levels which consist of microscopic, mesoscopic, and macroscopic models. In the microscopic model, a hand-tracking vision captures hand-skeletal data and finger joint angle features, enabling the estimation of physical hand postures. In the mesoscopic model, an egocentric vision with an eye tracker records to hand and eye movements, allowing for the symbolic representation of hand-eye coordination through hand gestures and visual attention focus during the test. An evaluation vision system employs color feature classification in the macroscopic model to determine whether the design matches the given task. Through the first eight designs of WAIS-IV BDT with two scenarios, the system successfully measures human behavior from the physical to the cognitive domain. The experiment involving eight healthy participants investigates the relationship between physical measurement and cognitive evaluation. Regression and correlation analyses between the dominant and non-dominant hands reveal that evaluation indices (task completion time, skewness-kurtosis of hand posture, attention to pattern and blocks) can indicate improvement during BDT. The outcomes of this study have significant implications for clinicians and researchers, providing valuable information that is typically unavailable in clinical settings. The proposed multiscopic CPSS framework holds promise for advancing independent rehabilitation practices. Code and datasets are available online at https://github.com/anom-tmu/bdt-multiscopic.

Abstract Dynamic locomotion is realized through a simultaneous integration of adaptability and optimality. This article proposes a neuro-cognitive model for a multi-legged locomotion robot that can seamlessly integrate multi-modal sensing, ecological perception, and cognition through the coordination of interoceptive and exteroceptive sensory information. Importantly, cognitive models can be discussed as micro-, meso-, and macro-scopic; these concepts correspond to sensing, perception, and cognition; and short-, medium-, and long-term adaptation (in terms of ecological psychology). The proposed neuro-cognitive model integrates these intelligent functions from a multi-scopic point of view. Macroscopic-level presents an attention mechanism with short-term adaptive locomotion control conducted by a lower-level sensorimotor coordination-based model. Macrosopic-level serves environmental cognitive map featuring higher-level behavior planning. Mesoscopic level shows integration between the microscopic and macroscopic approaches, enabling the model to reconstruct a map and conduct localization using bottom-up facial environmental information and top-down map information, generating intention towards the ultimate goal at the macroscopic level. The experiments demonstrated that adaptability and optimality of multi-legged locomotion could be achieved using the proposed multi-scale neuro-cognitive model, from short to long-term adaptation, with efficient computational usage. Future research directions can be implemented not only in robotics contexts but also in the context of interdisciplinary studies incorporating cognitive science and ecological psychology.

Most East Asian rehabilitation centers offer chopsticks manipulation tests (CMT). In addition to impaired hand function, approximately two-thirds of stroke survivors have visual impairment related to eye movement. This article investigates the significance of combining finger joint angle estimation and a visual attention measurement in CMT. We present a multiscopic framework that consists of microscopic, mesoscopic, and macroscopic levels. We develop a feature extraction technique to build the finger kinematic model at the microscopic level. At the mesoscopic level, we propose an active perception ability to detect the position and geometry of the finger on the chopsticks. The proposed framework estimates the proximal interphalangeal (PIP) joint angle on the index finger during CMT using fully connected cascade neural networks (FCC-NN). At the macroscopic level, we implement a cognitive ability by measuring visual attention during CMT. We further evaluate the proposed framework with a conventional test that counts the number of peanuts (NP) which are moved from one bowl to another using chopsticks within a particular time frame. We introduce three evaluation indices, namely joint angle estimation movement (JAEM), chopstick attention movement (CAM), and chopstick tip movement (CTM), by detecting the local minima and maxima of the time series data. According to the experiment results, the velocity of these three evaluation indices could indicate improvement in hand and eye function during CMT. We expect this study to benefit therapists and researchers by providing valuable information that is not accessible in the clinic. Code and datasets are available online at https://github.com/anom-tmu/cmt-attention.

This paper discusses a mechanism for integrating locomotion with cognition in robots. We demonstrate an attentional ability model that can dynamically change the focus of its perceptual area by integrating attention and perception to generate behavior. The proposed model considers both internal sensory information and also external sensory information. We also propose affordance detection that identifies different actions depending on the robot’s immediate possibilities. Attention is represented in a topological structure generated by a growing neural gas that uses 3D point-cloud data. When the robot faces an obstacle, the topological map density increases in the suspected obstacle area. From here, affordance information is processed directly into the behavior pattern generator, which comprises interconnections between motor and internal sensory neurons. The attention model increases the density associated with the suspected obstacle to produce a detailed representation of the obstacle. Then, the robot processes the cognitive information to enact a short-term adaptation to its locomotion by changing its swing pattern or movement plan. To test the effectiveness of the proposed model, it is implemented in a computer simulation and also in a medium-sized, four-legged robot. The experiments validate the advantages in three categories: (1) Development of attention model using topological structure, (2) Integration between attention and affordance in moving behavior, (3) Integration of exteroceptive sensory information to lower-level control of locomotion generator.

Hypertension is a global health problem and a leading factor in severe and life-threatening cardiovascular diseases (CVD) and stroke. The onset is dependent on individual lifestyle choices, and no single root cause of the condition exists. Various machine learning solutions are proposed for the early diagnosis of hypertension and its prediction, but they are based on standard guidelines and do not provide personalized solutions. Current models mainly rely on batch learning methods and do not readily learn the new incoming data. There is also a lack of an intelligent technique for handling anomalies in data, which leads to unreliable prediction results. In this paper, an integrated multi-agent-based hypertension risk prediction system is proposed that detects and computes missing values in the time series and provides personalized hypertension risk predictions. The proposed solution incorporates Gaussian mixture models for enhancing the input data, and an Online Infinite Echo State Gaussian Process (OIESGP) is used to obtain real-time prediction distribution of blood pressure. The prediction system readily absorbs new incoming data, and the model is updated to learn any new patterns in the data. The hypertension risk score is estimated using the Framingham hypertension risk estimator, and a 4-year hypertension risk is computed. The prediction performance of the proposed model is evaluated on blood pressure data gathered from the Malaysian population using mean absolute error, mean square error, and root-mean-square errors. The experimental results indicate that the proposed prediction model exhibits greater prediction accuracy than existing state-of-the-art online prediction methods.

With the advancement of artificial intelligence technologies in recent years, research on intelligent robots has progressed. Robots are required to understand human intentions and communicate more smoothly with humans. Since gestures can have a variety of meanings, gesture recognition is one of the essential issues in communication between robots and humans. In addition, robots need to learn new gestures as humans grow. Moreover, individual gestures vary. Because catastrophic forgetting occurs in training new data in traditional gesture recognition approaches, it is necessary to preserve the prepared data and combine it with further data to train the model from scratch. We propose a Multi-scopic Cognitive Memory System (MCMS) that mimics the lifelong learning process of humans and can continuously learn new gestures without forgetting previously learned gestures. The proposed system comprises a two-layer structure consisting of an episode memory layer and a semantic memory layer, with a topological map as its backbone. The system is designed with reference to conventional continuous learning systems in three ways: (i) using a dynamic architecture without setting the network size, (ii) adding regularization terms to constrain learning, and (iii) generating data from the network itself and performing relearning. The episode memory layer clusters the data and learns their spatiotemporal representation. The semantic memory layer generates a topological map based on task-related inputs and stores them as longer-term episode representations in the robot’s memory. In addition, to alleviate catastrophic forgetting, the memory replay function can reinforce memories autonomously. The proposed system could mitigate catastrophic forgetting and perform continuous learning by using both machine learning benchmark datasets and real-world data compared to conventional methods.

Activity recognition which aims to accurately distinguish human actions in complex environments plays a key role in human-robot/computer interaction. However, long-lasting and similar actions will cause poor feature sequence extraction and thus lead to a reduction of the recognition accuracy. We propose a novel discriminative deep model (D3D-LSTM) based on 3D-CNN and LSTM for both single-target and interaction action recognition to improve the spatiotemporal processing performance. Our models have several notable properties: 1) A real-time feature fusion method is used to obtain a more representative feature sequence through composition of local mixtures for enhancing the performance of discriminating similar actions; 2) We introduce an improved attention mechanism that focuses on each frame individually by assigning different weights in real-time; 3) An alternating optimization strategy is proposed for our model to obtain parameters with the best performance. Because the proposed D3D-LSTM model is efficient enough to be used as a detector that recognizes various activities, a Real-set database is collected to evaluate action recognition in complex real-world scenarios. For long-term relations, we update the present memory state via the weight-controlled attention module that enables the memory cell to store better long-term features. The densely connected bimodal modal makes local perceptrons of 3D-Conv motion-aware and stores better short-term features. The proposed D3D-LSTM model has been evaluated through a series of experiments on the Real-set and open-source datasets, i.e. SBU-Kinect and MSR-action-3D. Experimental results show that the proposed D3D-LSTM model achieves new state-of-the-art results, including pushing the average rate of the SBU-Kinect to 92.40% and the average rate of the MSR-action-3D to 95.40%.

The number of whiplash injuries in rear-end collisions is still high worldwide. Since the 1990s, many researchers have studied the mechanisms of injury occurrence in which were focused in the early stage of the rear impact. However, the current situation is that it has not led to a significant reduction in the number of injuries. The authors created a head-neck FEM model with muscles and accurately reproduced the head-neck behaviors of male and female volunteers in the rear impact tests. Then, using the same model, it was clarified that the difference in muscle strength is a large factor in the gender difference in frequency of whiplash injuries. In this paper, we focused on nerve damage, which is the cause of chronic whiplash, and examined the possibility of nerve root damage in the early stage of rear impact. It is generally known that living soft tissues such as nerves have viscoelastic behavior. Using animal axons and nerve root characteristics, it was found that a hyperbolic relationship was established between the velocity and the deformation under compression load on nerve root. The higher the loading rate, the stronger the locality at the compression site, and the greater the possibility of damage even if the deformation is small. Next, we studied the possibility of nerve root damage during a rear impact using a C4-C5 spinal unit FEM model including nerve root model (nerve-spinal unit model). First, in the same model, it was confirmed that there is no nerve root damage during a daily head extension. Then, using male head-neck behavior data in rear impact, it was confirmed that there is a possibility of nerve root damage when the compression speed of the vertebral body increases. Finally, we also analyzed using female data and found that there was a higher possibility of nerve root damage than in male.

There are currently many quadruped robots suited to a wide range of applications, but traversing some terrains, such as vertical ladders, remains an open challenge. There is still a need to develop adaptive robots that can walk and climb efficiently. This paper presents an adaptive quadruped robot that, by mimicking feline structure, supports several novel capabilities. We design a novel paw structure and several point-cloud-based sensory structures incorporating a quad-composite time-of-flight sensor and a dual-laser range finder. The proposed robot is equipped with physical and cognitive capabilities which include: 1) a dynamic-density topological map building with attention model, 2) affordance perception using the topological map, and 3) a neural-based locomotion model. The novel capabilities show strong integration between locomotion and internal–external sensory information, enabling short-term adaptations in response to environmental changes. The robot performed well in several situations: walking on natural terrain, walking with a leg malfunction, avoiding a sudden obstacle, climbing a vertical ladder. Further, we consider current problems and future development.

This paper proposes a topological clustering algorithm by integrating topological structure and information theoretic learning, i.e., correntropy, into adaptive resonance theory (ART). Specifically, the proposed algorithm utilizes the correntropy induced metric (CIM) for defining a similarity measure, a node insertion criterion, and an edge creation criterion. Other types of the ART-based topological clustering algorithms have been developed, however, these algorithms have various drawbacks such as a large number of parameters, sensitivity to noisy data. Moreover, generated topological networks cannot represent the distribution of data. In contrast, the proposed algorithm realizes a stable computation and reduces the number of parameters compared to existing algorithms. Furthermore, improving the ability to express the data structure more appropriately by the topological network, a mechanism that adaptively controls the node insertion criterion is introduced to the proposed algorithm. The experimental results showed that the proposed algorithm has superior performance with respect to the self-organizing and the classification abilities compared with the state-of-the-art topological clustering algorithms.

Global path planning is often used for initial path planning, as it allows robots to search for the feasible paths. However, it is very difficult to perform global path planning in a dynamic environment because it is hard to predict the behavior of moving obstacles. In this paper, we propose a method to perform a global path plan by using a spatiotemporal graph for environments with moving obstacles that perform predictable behavior. We provide a waiting path in a spatiotemporal graph and setting the waiting cost on it. Then, we show that it is possible to adjust the robot's behavior by using A* algorithm from the balance of moving cost and waiting cost. We also show that the global path from the current position to the goal position can be easily updated by using a spatiotemporal graph, even when moving obstacles change from the scheduled behavior. In order to validate the effectiveness of the proposed method, we apply it for return-path planning of an autonomous tour guide robot in a virtual museum. Finally, we show the simulation results and discuss the applicability of the proposed method in a dynamic environment.

From worldwide statistical data of whiplash injuries in the rear-end impact accidents, females have been found to be at a higher risk than males and the reason for the same is an ongoing research topic. First, we created finite element model of the head and neck of a 50th %ile Japanese adult female (JAF50) by scaling an existing 50th %ile (JAM50) male model. The activity level of neck muscles before a rear-end impact was determined using the L36 of DOE method, which held a neutral posture of the head with the relaxed muscles under 1G condition. Subsequently, the head and neck behaviors of both the models were validated using the volunteer's rear-end impact test data. Comparing both the results using a new index (S - θmax), which measured the S-shape deformation of the neck during rear-end impacts, the model results were found to have the same trends as that of the volunteer's test data. JAF50 had a higher (S - θmax) value than JAM50, which means that a female neck undergoes larger S-shaped mode than a male. This index value difference between the genders was due to the difference in their muscle forces while maintaining a neutral posture of the head. This difference in the muscle forces, however, was not proportional to only the head mass ratio of the two genders. As a result of the investigation of the underlying reason, the neck muscle force ratio of each gender at the neutral position was roughly related to mass ratio × neck column cross-sectional area ratio × neck length ratio. If the ratio (female to male) of neck circumferences was 3: 4, the female muscle force could be nearly twice less than that of a male. We also estimated the muscle activity for each gender assuming the same (S - θmax) = 15 ° at the rear-end impact of Δ V = 5 km/h. It was found that a female's muscle activity is about twice that of a males for the same rear-end impact conditions.

This paper presents a robot partner development platform based on smart devices. Humans communicate with others based on the basic motivations of human cooperation and have communicative motives based on social attributes. Understanding and applying these communicative motives become important in the development of socially-embedded robot partners. Therefore, it is becoming more important to develop robots that can be applied according to needs while taking these human communication elements into consideration. The role of a robot partner is more important in not only on the industrial sector but also in households. However, it seems that it will take time to disseminate robots. In the field of service robots, the development of robots according to various needs is important and the system integration of hardware and software becomes crucial. Therefore, in this paper, we propose a robot partner development platform for human-robot interaction. Firstly, we propose a modularized architecture of robot partners using a smart device to realize a flexible update based on the re-usability of hardware and software modules. In addition, we show examples of implementing a robot system using the proposed architecture. Next, we focus on the development of various robots using the modular robot partner system. Finally, we discuss the effectiveness of the proposed robot partner system through social implementation and experiments.

In this paper, a new approach is proposed for mobile robot localization and hybrid map building simultaneously without using any odometry hardware system. The proposed method termed as Genetic Bayesian ARAM which comprises two main components: 1) Steady state genetic algorithm (SSGA) for self-localization and occupancy grid map building; 2) Bayesian Adaptive Resonance Associative Memory (ARAM) for online topological map building. The model of the explored environment is formed as a hybrid representation, both topological and grid-based, and it is incrementally constructed during the exploration process. During occupancy map building, robot estimated self-position is updated by SSGA. At the same time, robot estimated self position is transmit to Bayesian ARAM for topological map building and localization. The effectiveness of our proposed approach is validated by a number of standardized benchmark datasets and real experimental results carried on mobile robot. Benchmark datasets are used to verify the proposed method capable of generating topological map in different environment conditions. Real robot experiment is to verify the proposed method can be implemented in real world.

The plateau stress of aluminum (Al) foam is an important mechanical parameter that is closely related to its energy absorptivity. An estimation method for plateau stress was proposed by simply assuming that when the mean true compressive stress on a maximum-porosity cross section perpendicular to the direction of compressive loading reaches the critical value, the nominal compressive stress becomes equal to the plateau stress. In this study, using commercial-purity A1050 Al, A6061 Al alloy and ADC12 Al alloy die castings containing a large amount of gases, uniform Al foam and two types of functionally graded (FG) Al foam were fabricated by the friction stir welding (FSW) route precursor process. The local porosity distributions for the fabricated Al foams were obtained from the observation of X-ray computed tomography (CT) images, then drop weight impact tests were carried out to evaluate the plateau stresses. Through the comparison of the test results with the results obtained by the estimation method, the applicability of the estimation method for plateau stress to impact compression tests on uniform aluminum Al foam and each foam layer of FG Al foam was examined. It was shown that the plateau stresses can be evaluated approximately within ±20% error when the proof stress is employed as the critical value.

This paper discusses associative learning of a partner robots through interaction with people. Human interaction based on gestures is very important to realize the natural communication. The meaning of gestures can be understood through the actual interaction with a human and the imitation of a human. Therefore, we propose a method for associative learning based on imitation and conversation to realize the natural communication. Steady-state genetic algorithms are applied for detecting human face and objects in image processing. Spiking neural networks are applied for memorizing spatio-temporal patterns of human hand motions, and relationship among perceptual information. Furthermore, we conduct several experiments of the partner robot on the interaction based on imitation and conversation with people. The experimental results show that the proposed method can refine the relationship among the perceptual information, and can reflect the updated relationship to the natural communication with a human.