Volume 8, Issue 6

A robot development has spread widely for various purposes. It is difficult to create a control program for an autonomous mobile robot manually. Therefore, an automatic design of the control program for an autonomous mobile robot is proposed in this research. The autonomous mobile robot is created with LEGO MINDSTORMS EV3, and the control program for the au- tonomous mobile robot is designed using Grammatical Evolution (GE). Grammatical Evolution (GE), which is one of the evolutionary computations, is designed to generate a program or a program fragment satisfying the design objective. PyBullet is used with GE to simulate the behavior of the robot. A robot traveling along a trajectory was considered as an example. GE can generate the control program of the robot behavior of a robot vehicle traveling along a trajectory. The computer simulation reveals the robot can travel along a designated line. Since there is a reality gap between the simulator and the real environment, the parameters of the vehicle robot such as produced power and sensor sensitivity are calibrated to reduce the gap. Comparison of the computer simulation and the experimental result shows that the reproducibility of the vehicle trajectory in the real environment is high.

Read more…

A robot development has spread widely for various purposes. It is difficult to create a control program for an autonomous mobile robot manually. Therefore, an automatic design of the control program for an autonomous mobile robot is proposed in this research. The autonomous mobile robot is created with LEGO MINDSTORMS EV3, and the control program for the au- tonomous mobile robot is designed using Grammatical Evolution (GE). Grammatical Evolution (GE), which is one of the evolutionary computations, is designed to generate a program or a program fragment satisfying the design objective. PyBullet is used with GE to simulate the behavior of the robot. A robot traveling along a trajectory was considered as an example. GE can generate the control program of the robot behavior of a robot vehicle traveling along a trajectory. The computer simulation reveals the robot can travel along a designated line. Since there is a reality gap between the simulator and the real environment, the parameters of the vehicle robot such as produced power and sensor sensitivity are calibrated to reduce the gap. Comparison of the computer simulation and the experimental result shows that the reproducibility of the vehicle trajectory in the real environment is high.

Read more…

An ensemble method is a Machine Learning (ML) algorithm that aggregates the predictions of multiple estimators or models. The purpose of an ensemble module is to provide better predictive performance than any single contributing model. This can be achieved by producing a predictive model with reduced variance using bagging, and bias using boosting.
The Tree-Based Ensemble Models with Decision Tree (DT) as base model is the most frequently used. On the other hand, there are some individual Machine Learning algorithms that can provide more competitive predictive power to the ensemble models. It is a problem, and this issue is addressed here. This work has two parts. The first one presents a Projective Decision Tree (PA) based on purity measure. Next node criterion (CNN) is also used for node decision making. In the second part, two sets of algorithms for predictive performance are presented. The Tree-Based Ensemble model includes bagging and boosting for homogeneous learners and a set of known individual algorithms. Comparison of two sets is performed for accuracy. Furthermore, the changes of bagging and boosting ensemble performance under various hyperparameters are also investigated. The datasets used are the sonar and the Breast Cancer Wisconsin (BCWD) from UCI site. Promising results of the proposed models are accomplished.

Read more…

Daily life is characterized by a great explosion of abundance of information available on the internet and social media. Smart technology has radically changed our lives, giving a leading role to social media for communication, advertising, information and exchange of opinions. Managing this huge amount of data by humans is an almost impossible task. Adequacy of summarizing texts is therefore urgently needed, in order to offer people knowledge and information avoiding time-consuming procedures. Various text summarization techniques are already widely used. Artificial intelligence techniques for automated text summarization are a major undertaking. Due to the recent development of neural networks and deep learning models like Transformers, we can create more efficient summaries. This paper reviews text summarisation approaches on social media and introduces our approach towards a summarization system using transformers.

Read more…

In the context of adopting cloud Infrastructure-as-a-Service (IaaS), prospective consumers need to consider a wide array of both business and technical factors associated with the service. The development of an intelligent tool to aid in the assessment of IaaS offerings is highly desirable. However, the creation of such a tool requires a robust foundation of domain knowledge. Thus, the focus of this paper is to introduce an ontology specifically designed to characterize IaaSs from the consumer’s perspective, enabling informed decision-making. The ontology additionally serves two purposes of other relevant parties besides the consumers. Firstly, it empowers IaaS providers to better tailor their services to align with consumer expectations, thereby enhancing their competitiveness. Additionally, IaaS partners can play a pivotal role in supporting both consumers and providers by understanding the protocol outlined in the ontology that governs interactions between the two parties. By applying principles of ontological engineering, this study meticulously examined the various topics related to IaaS as delineated in existing cloud taxonomies. These topics were subsequently transformed into a standardized representation and seamlessly integrated through a binary integration approach. This process resulted in the creation of a comprehensive and cohesive ontology that maintains semantic consistency. Leveraging Protégé, this study successfully constructed the resultant ontology, comprising a total of 340 distinct classes. The study evaluated the syntactic, semantic, and practical aspects of the ontology against a worldwide prominent IaaS. The results showed that the proposed ontology was syntactically and semantically consistent. Furthermore, the ontology successfully enabled not only the assessment of a real leading IaaS but also queries to support automation tool development.

Read more…

Mental stress assessment remains riddled with biases caused by subjective reports and individual differences across societal backgrounds. To objectively determine the presence or absence of mental stress, there is a need to move away from the traditional subjective methods of self-report questionnaires and interviews. Previously, it has been evidence that EEG Oscillations can discriminate mental states, for instance, stressed and non-stressed. However, it is still not clear in which range of EEG oscillations the neural activities are associated with the mental states. This paper presents a wavelet-based EEG feature extraction method for the classification of mental stress using machine learning classifiers. An EEG dataset of 22 participants was used to test the performance of the proposed wavelet-based feature extraction method. The dataset includes both stress and control conditions, and the stress condition has multiple levels of stress, starting from low, mild, and high stress. The Daubechies mother wavelet of the fourth order was used to separate the EEG oscillations into 7 levels for the extraction of the absolute powers. Whereas Fast Fourier Transform were implemented to obtain the average power of the oscillations. The features were then used in support vector machine, decision tree, linear discriminant analysis and artificial neural network classifiers. A comparison between the classifiers using average power, absolute power, and a combination of both is provided. The EEG alpha, theta, and beta frequency bands showed promising results for the classification of mental stress vs. control conditions by achieving an average accuracy of 95% using the decision tree. The results of the proposed method suggest the potential use of wavelet analysis for mental stress detection despite FFT performing better. The proposed method has the potential to be used in Computer-Aided Diagnosis (CAD) systems for mental stress assessment in the future alongside the discovery of significant wave bands in relation to mental stress detection.

Read more…

Liver cancer is a major contributor to cancer-related mortality both in the United States and worldwide. A range of liver diseases, such as chronic liver disease, liver cirrhosis, hepatitis, and liver cancer, play a role in this statistic. Hepatitis, in particular, is the main culprit behind liver cancer. As a consequence, it is decisive to investigate the correlation between hepatitis and symptoms using statistic inspection. In this study, we inspect 155 patient data possessed by CARNEGIE-MELLON UNIVERSITY in 1988 to prognosticate whether an individual died from liver disease using supervised machine learning models for category and connection rules based on 20 different symptom attributes. We compare J48 (Gain Ratio) and CART (Classification and Regression Tree), two decision tree classification algorithms elaborate from ID3 (Iterative Dichotomiser 3), with the Gini index in a Java environment. The data is preprocessed through normalization. Our study demonstrates that J48 outperforms CART, with an average accuracy rate of nearly 87% for the complete specimen, cross-validation, and 66% training data. However, CART has the supreme accurate rate in all samples, with an accuracy rate of 90.3232%. Furthermore, our research indicates that removing the conjunction attribute of the Apriori algorithm does not impact the results. This research showcases the potential for physician and researchers to apply brief machine learning device to attain accurate outcomes and develop treatments based on symptoms.

Read more…

Many prosthetic hands are focused on appearance and grip strength, however, gestures are also one of the performances that users need for communicating with others as body language to express their feeling and intention. For this paper, the initial prototype of the gesturing robotics hand is presented by using multiple types of actuators concept to maintain its appearance while the number of degrees of freedom (DOFs) is increased. The gesture performance of this robotic hand is improved by designing 2 DOFs in each finger; therefore, the entire hand has 15 joints, 10 DOFs, and one controllable wrist joint. In the detail of the design, all actuating mechanisms of 3 joints for each finger are specifically designed to maintain the human limb appearance. The Distal Phalange joint (DIP) used a linkage mechanism to acquire the Proximal Phalange joint’s (PIP) movement by a 1:0.961 ratio due to appearance designed. The PIP joint is built with a cable-drive mechanism powered by a digital servo motor installed in the forearm. The Metacarpal joint (MCP) is driven by a micro linear actuator in the hand palm. A micro linear actuator was also selected to drive the wrist of the robot. The hand can perform 10 hand gestures follow common emoji hand gestures and holding 12 objects in the hand. The design of this bio inspired robot hand can be downloaded for educational purpose.

Read more…

In this paper, we have developed a gas injection-type prefabricated lifting device based on location tracking to efficiently lift the human body in the event of an accident that occurs underwater on the sea or land. The efficiency of the lifting system is very important to ensure the golden time of the rescue and the safety of divers in the event of casualties underwater. Divers performing underwater safety rescue operations must endure up to 30 minutes with two air vents, and always consider the safety accident environment due to difficulty in securing visibility or high flow rates due to underwater turbidity. Particularly, there are many cases where life is threatened by hypothermia in the water. Therefore, both divers and the deceased need location tracking connected to the lifting device, and a fast and efficient lifting system was studied in underwater activities. The monitoring device uses a communication speed of 115.2 kbps from the sensor to the monitoring, and a communication speed of 2.4 kbps from the controller to the receiving unit. The gas injection-type prefabricated lifting device with a high elastic structure is lightweight and portable, and which consists of a baggy bag with minimal components to increase usage and work efficiency based on the instinctive behavior of divers. Accordingly, the entrance element design combining a bow and hinge that maintains a moment of force with TPU-based materials, a balanced design using weight balancing technology of a network structure, an SMB linkage design that induces water surface rise through gas injection, and an underwater experiment.

Read more…

The advent of laparoscopic surgery has increased the need to incorporate simulator-based training into traditional training programs to improve resident training and feedback. However, current training methods rely on expert surgeons to evaluate the dexterity of trainees, a time-consuming and subjective process. Through this research, we aim to extend the use of object detection in laparoscopic training by detecting and tracking surgical tools and objects. In this project, we trained YOLOv7 object detection neural networks on Fundamentals of Laparoscopic Surgery pattern-cutting exercise videos using a trainable bag of freebies. Experiments show that YOLOv7 has a mAP score of 95.2, 95.3 precision, 94.1 Recall, and 78 accuracy for bounding boxes on a limited-size training dataset. This research clearly demonstrates the potential of using YOLOv7 as a single-stage real-time object detector in automated tool motion analysis for the assessment of the resident’s performance during training.

Read more…

A reliable healthcare system ensures that the population has access to top-notch medical ser- vices, ultimately enhancing their overall health most efficiently. At times, data are not secured or handled appropriately. Addressing these concerns, blockchain technology is projected to bring about a substantial revolution in the medical industry by assuring the confidentiality of electronic health information. This research not only seeks to rectify the shortcomings in Bangladesh’s existing health system but also explores the potential of blockchain technology’s decentralized database to fortify the entire healthcare framework. More importantly, it show- cases a web-based application, particularly a medical history card that displays a patient’s details, diagnoses, vaccines, medication records, investigation background, familial information, blood donation history, and many additional information starting from birth. Alongside, the paper emphasizes the transformative impact of implementing blockchain technology in the healthcare sector, paving the way for a more secure and efficient healthcare ecosystem. All in all, the array of medical information captured within the pack face of a single card could hasten medical decisions and ensure the effectiveness of any treatment.

Read more…

Cloud computing has emerged as a pivotal component of contemporary IT systems, affording organizations the agility and scalability required to meet the ever-changing demands of business. However, this technological evolution has introduced a new era of cybersecurity challenges, as attackers employ increasingly sophisticated strategies to breach cloud networks. Such breaches can have far-reaching consequences, including data loss, financial repercussions, reputational damage, and legal liabilities. In response to these challenges, developing a robust security framework is imperative for effectively safeguarding cloud infrastructure. This paper proposes a novel Hypervisor-based Virtual Machine Introspection (HVMI) for real-time detection and runtime forensic analysis of cyberattacks targeting cloud platforms. The framework proposed comprises several essential components, including a forensic application empowered by Virtual Machine Introspection (VMI) for real-time memory analysis, a centralized Cloud Forensic Tool (CFT) portal for streamlined incident management, and a data transmission and integration web service. Notably, this framework is founded upon a commitment to continuous optimization and enhancement. This iterative process is facilitated through a collaboration approach. It involves fine-tuning various aspects of the framework, such as adjusting settings for VMI, refining criteria for classifying incidents, and updating security controls. Enhancing the forensic application represents a proactive measure aimed at improving the efficiency and effectiveness of VMI and forensic analysis capabilities. The iterative refinement process integrates incident analysis, threat intelligence infusion, and collaborative efforts to adapt to emerging challenges. This dynamic approach fosters a flexible security posture capable of detecting, analyzing, and responding to emerging attacks within cloud platforms. In summary, the proposed framework embodies a comprehensive approach to cloud security, integrating advanced technology with continuous refinement to protect cloud infrastructure, mitigate risks, and navigate the ever-evolving cybersecurity threat landscape effectively.

Read more…

We present an inverted pendulum design using readily available V-slot rail components and 3D printing to construct custom parts. To enable the examination of different pendulum characteristics, we constructed three pendulum poles of different lengths. We implemented a brake mechanism to modify sliding friction resistance and built a paddle that can be attached to the ends of the pendulum poles. A testing rig was also developed to consistently apply disturbances by tapping the pendulum pole, characterizing balancing performance. We perform a comprehensive analysis of the behavior and control of the pendulum. This begins by considering its dynamics, including the nonlinear differential equation that describes the system, its linearization, and its representation in the s-domain. The primary focus of this work is the development of two distinct control modes for the pendulum: a velocity control mode, designed to balance the pendulum while the cart is in motion, and a position control mode, aimed at maintaining the pendulum cart at a specific location. For this, we derived two different state space models: one for implementing the velocity control mode and another for the position control mode. In the position control mode, integral action applied to the cart position ensures that the inverted pendulum remains balanced and maintains its desired position on the rail. For both models, linear observer-based state feedback controllers were implemented. The control laws are designed as linear quadratic regulators (LQR), and the systems are simulated in MATLAB. To actuate the physical pendulum system, a stepper motor was used, and its controller was assembled in a DIN rail panel to simplify the integration of all necessary components. We examined how the optimized performance, achieved with the medium-length pendulum pole, translates to poles of other lengths. Our findings reveal distinct behavioral differences between the control modes.

Read more…

Cybercriminal exploits integrity, confidentiality, and availability of information resources. Cyberattacks are typically invisible to the naked eye, even though they target a wide range of our digital assets, such as internet-connected smart devices, computers, and networking devices. Implementing network anomaly detection proves to be an effective method for identifying these malicious activities. The traditional anomaly detection model cannot detect zero-day attacks. Hence, the implementation of the artificial intelligence method overcomes those problems. A specialized model, known as a recurrent neural network (RNN), is specifically crafted to identify and utilize sequential data patterns to forecast upcoming scenarios. The random selection of hyperparameters does not provide an efficient result for the selected dataset. We examined seven distinct optimizers: Nadam, Adam, RMSprop, Adamax, SGD, Adagrad, and Ftrl, with variations in values of batch size, epochs, and the data split ratio. Our goal is to optimize the performance of the bidirectional long short-term memory (Bi-LSTM) anomaly detection model. This optimization resulted in an exceptional network anomaly detection accuracy of 98.52% on the binary NSL-KDD dataset. Sampling techniques deal with the data imbalance problem. Random under-sampling, which involved removing data from the majority classes to create a smaller dataset, was less efficient for deep learning models. In contrast, the Synthetic Minority Oversampling Technique (SMOTE) successfully generated random data related to the minority class, resulting in a balanced NSL-KDD multiclass dataset with 99.83% Bi-LSTM model detection accuracy. Our analysis discovered that our Bidirectional LSTM anomaly detection model outperformed existing anomaly detection models compared to the performance metrics, including precision, f1-score, and accuracy.

Read more…