The presented research focuses on Hand Gesture Recognition (HGR) utilizing Surface-Electromyogram (sEMG) signals. This is due to its unique potential for decoding wearable data to interpret human intent for immersion in Mixed Reality (MR) environments. The existing solutions so far rely on complicated and heavy-weighted Deep Neural Networks (DNNs), which have restricted practical application in low-power and resource-constrained wearable systems. In this work, we propose a light-weight hybrid architecture ( $$\text {HDCAM}$$ ) based on Convolutional Neural Network (CNN) and attention mechanism to effectively extract local and global representations of the input. The proposed $$\text {HDCAM}$$ model with 58, 441 parameters reached a new state-of-the-art (SOTA) performance with $$83.54\%$$ and $$82.86\%$$ accuracy on window sizes of 300 ms and 200 ms for classifying 17 hand gestures. The number of parameters to train the proposed $$\text {HDCAM}$$ architecture is $$18.87 \times $$ less than its previous SOTA counterpart. Furthermore, the model is trained based on a hybrid loss function consisting of two-fold: (i) Cross Entropy (CE) loss which focuses on identifying the helpful features to perform the classification objective, and (ii) Supervised Contrastive (SC) loss which assists to learn more robust and generic features by minimizing the ratio of intra-class to inter-class similarity.
Ultrasound (US) is the most widely used medical imaging modality due to its low cost, portability, real time imaging ability and use of non-ionizing radiation. However, unlike other imaging modalities such as CT or MRI, it is a heavily operator dependent, requiring trained expertise to leverage these benefits. Recently there has been an explosion of interest in AI across the medical community and many are turning to the growing trend of deep learning (DL) models to assist in diagnosis. However, due to possible differences in training and deployment, model performance suffers which can lead to misdiagnosis and operator hesitancy. This issue is known as dataset shift. Two aims to address dataset shift were proposed. The first was to quantify how US operator skill and hardware affects acquired images. The second was to use this skill quantification method to screen and match data to deep learning models to improve performance. A CAE Healthcare BLUE phantom with mock lesions was scanned by three operators using three different US systems (Siemens S3000, Clarius L15, and Ultrasonix SonixTouch) producing 39013 images. DL models were trained on a specific set to classify the presence of a simulated tumour and tested with data from differing sets. Principle Component Analysis (PCA) for dimension reduction was applied, then K-Means clustering was used to separate images generated by operator and hardware into clusters. This clustering algorithm was then used to screen incoming images during deployment to best match input to an appropriate DL model which is trained specifically to classify that type of operator or hardware. Results showed a noticeable difference when models were given data from differing datasets with the largest accuracy drop being 81.26% to 31.26%. Overall, operator differences more significantly affected DL model performance. Clustering models had much higher success separating hardware data compared to operator data. The proposed method reflects this result with a much higher accuracy across the hardware test set compared to the operator data.
Fast-paced and ever-growing advances in Artificial Intelligence (AI) and Deep Neural Network (DNN) models have initiated works on autonomous monitoring/screening systems to assess individuals’ cognitive state. In conventional cognitive assessment systems, a physician evaluates the mental abilities of the brain by rating the patient’s numerical, verbal, and logical responses. Development of an autonomous cognitive assessment system that assist the physician is both of significant importance and a critically challenging task. As a first step towards achieving this objective, in the paper an Automated Virtual Cognitive Assessment ( $$\text {AVCA}$$ ) framework is proposed that integrates Natural Language Processing (NLP) and hand gesture recognition techniques. The proposed $$\text {AVCA}$$ framework provides individual scores in the seven major cognitive domains, i.e., orientation, attention, language, contractual ability, memory, calculation, and reasoning. More specifically, the $$\text {AVCA}$$ framework is an autonomous cognitive assessment system that receives audio and video signals in a real-time fashion, and performs semantic and synthetic analysis using NLP techniques and DNN models. Real-time video processing engines of the $$\text {AVCA}$$ monitors hand motions and facilitate simpler engagement for visual evaluation. Additionally, we propose an efficient model to facilitate human-machine interactions from speech recognition to text classification. In particular, an unsupervised contrastive learning framework is proposed using Bidirectional Encoder Representations from Transformers (BERT) that outperforms its state-of-the-art unsupervised counterparts achieving an average of 77.84% Sparsman’s correlation on standard Semantic Textual Similarity (STS) tasks.
The objective of this paper is to formalize the stress contagion protocols in first responder teams which respond to life-threatening crises on a daily basis. Such teams include both human and autonomous machine teammates. While the stress contagion is an attribute of human relations, collaboration between the humans under stress and, and the stressed humans and machines, impacts the mission performance. Thus, it is important to model this process and use the modeling to train the first responder teams. This paper proposes a framework for modeling the human-machine communication protocols using the advances in computational multivalued logic. We also separate the formalization of the protocol from the choice of an appropriate model. This separation allows to mitigate the constraints posed by the limited real-world initial data of absence of thereof. The proposed formalization based on multivalued logic is well-supported by the computational and benchmarking tools.
Developing a trustworthy biometric-enabled autonomous and semi-autonomous system involves computing the amount of bias and fairness of the data upon which the decisions are made. In this paper, we evaluate the performance of a system that performs human users’ face identification in terms of hit and miss rates. We also assess how the usage of data for training biometric systems can lead to unfairness, especially in the context of evolving biometric traits in times of pandemics. In particular, we contrast the performance of normal “unmasked” facial images with their “masked” counterparts. The SpeakingFace and Thermal-Mask dataset is used for assessment, where the Thermal-Mask dataset consists of synthetically applied masks on both the thermal (IR) and visual (RGB) domains. The comparative experiment assesses how fairness metrics such as demographic parity difference (DPD) and equalized odds difference (EOD) can be applied to both masked and normal facial images across the thermal and visual domains. For age, gender, and ethnicity demographic groups, the DPD approaches 0.10%, 0.01%, and 0.01% when the hit rate and miss rate for facial identification approach 100% and 0%, respectively. For masked-thermal facial identification using a simple 2-block convolutional neural network, we obtain a hit rate of 69.96% whereas an 85.82% hit rate is reported for the unmasked alternative. The corresponding age DPD for such hit rate is 49.06% and 34.02% for their respectively masked-thermal and normal-thermal images. We conclude that datasets with inherent biases influence the fairness of a biometric system. The bias manifests itself across the demographic groups (age, gender, and ethnicity), the spectral domain (IR or RGB), and whether synthetic procedures are applied to generate new images. In particular, the error rates among masked facial images are consistently above the unmasked alternative for both visual and thermal domains.
It has been well understood that fake news recognition is a persistent challenge to cognitive computing and autonomous systems in general, and to Artificial Intelligence (AI), machine knowledge learning, and computational linguistics in particular. This work develops an Autonomous Fake News Recognition (AFNR) system by cognitive computing theories underpinned by Intelligent Mathematics (IM) such as concept algebra and semantic algebra. A training-free methodology and a formal algorithm for Differential Sematic Analysis (DSA) are designed in Real-Time Process Algebra (RTPA) and implemented in MATLAB. The AFNR system is implemented in the Anaconda environment with Python, the natural language toolkit (NLTK), and an English parser – Spacy. Compared to the classical data-driven neural network methodologies, AFNR and DSA have demonstrated a significant improvement against the level of accuracy over the randomly selected and large-scale benchmark of a fake news database. The DSA methodology for fake news recognition has enabled autonomous machine knowledge learning and semantic comprehension towards differential and robust semantic analyses for fake news in natural languages. The AFNR system has reached an accuracy level of 70.1%, which over performs the top ranked teams in DataCup’19 with the highest reported accuracy of 55.0%.
The paper initially addresses Malaysia’s readiness concerning the adoption of cloud computing services and provides insights into the country’s internet infrastructure and internet users, fixed and mobile broadband deployment, and usage of international bandwidth for improvising cloud services. In addition, the paper investigates the scenario around policy areas, including privacy and security laws, being the chief contributors for the adoption of cloud services across regions and borders. The objective behind conducting the research is to validate the current scenario in Malaysia for the cloud service adoption by healthcare and education segment and identifying the consumer service attribute preferences. By performing the conjoint analysis the research would help in identifying the determinants and prioritizing them based on user preferences with prior use of Multinomial Logit choice model (MNL) and Max Likelihood function. A statistical discussion is also presented concerning the current global market scenario and scenario in the Asia Pacific across the verticals.
Due to its rapidly advancing spread, the world is still reeling from COVID-19 (coronavirus 2019), which is categorized as a highly infectious disease. An early diagnosis is very critical in treating COVID-19 patients due to its lethal implications. However, the shortage of X-ray machines has resulted in life-threatening conditions and delays in diagnosis, increasing the number of deaths around the world. Therefore, in order to avoid such fatalities, COVID-19 has to be detected earlier and diagnosed faster using an intelligent computer-aided diagnosis system than with traditional screening programs. We present a novel framework for COVID-19 image categorization in this article that utilizes deep learning (DL) and bio-inspired optimization techniques. A bio-heuristic optimizer algorithm MoFAL is utilized as a feature selector to decrease the dimensionality of the image representation and increase the accuracy of the classification by ensuring that only the most essential selected features are used. Furthermore, the feature extraction is realized using the MobileNetV3 DL model. The experimental results deduced indicate that our proposed approach drastically improves performance in terms of classification accuracy and reduction in dimensions reflected during the period of feature extraction and its phases of selection. We propose that our COVID-Classifier can be deployed in conjunction with other tests for optimal allocation of hospital resources by rapid triage of non-COVID-19 cases.
With the increase in the number of processing elements (PEs) in modern highly integrated parallel systems, there has been a growing importance for designing an efficient self-restructuring method to automatically tolerate faulty PEs. In this paper, we present a self-restructuring method for mesh-connected processor arrays with spares on the orthogonal sides, based on the cooperation of the redundancy and degradation approaches. The redundancy approach replaces faulty PEs with spares, while the degradation approach deletes rows and/or columns of the arrays. First, we formalize the spare assignment problem in the redundancy approach as a matching problem in graph theory. Then, if no matching, i.e., no valid spare assignment, is found for an array with faulty PEs, rows and/or columns are deleted from the array so that a matching is successfully found for a degraded subarray. Finally, hardware circuits to realize the above process are presented. This leads to the realization of degradable self-restructuring of processor arrays, and implies that the proposed method is useful in enhancing especially the run-time reliability and availability of processor arrays in mission critical applications where first self-reconfiguration is required without an external host computer.
Recently, AutoEmbedder has been introduced that implements DNN classifiers as an embedding system. Although the AutoEmbedder generates clusterable embeddings, it is trained in a supervised approach. Therefore, in this paper, we introduce an unsupervised AutoEmbedder (UAutoEmbedder) that is trained in an unsupervised fashion. Through rigorous research effort, we discovered that AutoEmbedder is an ideal architecture for data augmentation dependent unsupervised learning. Hence, we propose a random sampling scheme and perform image data augmentation to train the AutoEmbedder system in an unsupervised manner. Furthermore, we reform the training procedure of the AutoEmbedder that properly utilizes the unsupervised learning strategy. We evaluate the model by applying different datasets and conduct an in-depth study of the generated results. From the comprehensive evaluations, we perceive that UAutoEmbedder has enormous opportunities in unsupervised learning and image data augmentation.
This paper proposes a new Structural Composite Feature Triangulation (SCFT) framework for Visual Object Search. Visual Object Search methods often use feature matching approaches which adopt Bag-of-Words (BoW) image representation and employ Geometric Verification (GV) to check the geometric consistency of the putative matching feature pairs between the query and the database image. However, these approaches are unable to handle multiple object instances in an image and they do not utilize the joint local structural information of the matching feature pairs to perform visual search. In view of this, this paper proposes the SCFT method to address these shortcomings. Its key contributions center on the formulation and proposal of a Structural Composite Feature Triangulation method to mine and detect composite structures from Delaunay Triangulation. The composite structures perform detection and localization of multiple object instances within an image. Experimental results on the Belgalogos dataset [1] and the Traffic sign dataset [2] demonstrate that the proposed method can detect multiple object instances, and the top retrieved images show that the proposed method is able to effectively retrieve the relevant images. The results show that SCFT achieves around 4% increase in mean Average Precision (mAP) when compared with that obtained by state-of the-art methods for the Belgalogos dataset.
We propose a scheme for generating a weakly chordal graph from a randomly generated input graph, G = (V, E). We reduce G to a chordal graph H by adding fill-edges, using the minimum vertex degree heuristic. Since H is necessarily a weakly chordal graph, we use an algorithm for deleting edges from a weakly chordal graph that preserves the weak chordality property of H. The edges that are candidates for deletion are the fill-edges that were inserted into G. In order to delete a maximal number of fill-edges, we maintain these in a queue. A fill-edge is removed from the front of the queue, which we then try to delete from H. If this violates the weak chordality property of H, we reinsert this edge at the back of the queue. This loop continues till no more fill-edges can be removed from H. Operationally, we implement this by defining a deletion round as one in which the edge at the back of the queue is at the front.We stop when the size of the queue does not change over two successive deletion rounds and output H.
This paper aims at the design and development of two hybrid nature inspired algorithms based on Grey Wolf Optimizer and Whale Optimization Algorithm (GWOWOA) and Binary Bat Optimization Algorithm and Particle Swarm Optimization Algorithm (BATPSO). Hybridization is a useful method to enhance the performance of these algorithms. The GWO algorithm is easy to fall into local optimum especially when it is used in the high-dimensional data. WOA algorithm experiences relatively low convergence precision and poor rate of convergence when it is applied in complex optimization problems. In this paper we embed bubble-net foraging method in WOA with that of prey encircling method in GWO. It has been observed experimentally that than just updating position vectors with respect to the three best fitness solutions, we were able to achieve faster convergence and better global optimum in most cases. In BATPSO, both algorithms are integrated and run in parallel and they perform a comparison between both minimum fitness function at each iteration. According to the observation, this greedy search algorithm in BAT optimization works best in higher values; however, finds it difficult in finding the global minimum as it reaches lower values; especially fractional fitness value. PSO is based on element-wise pos[i,j] search and updating the velocity to converge to a global minimum. The proposed hybrid algorithms are bench-marked using a set of 23 classical benchmark functions employed to test different characteristics of hybrid optimizers. The paper also performs solving two classical engineering design problems - Cantilever beam design and the multiple disc clutch brake problem. The results of the fitness functions prove that the proposed hybrid algorithms are able to produce better or very competitive output with respect to improved exploration, local optima avoidance, exploitation and convergence. All these hybrid algorithms find superior optimal designs for quintessential engineering problems engaged, showcasing that these algorithms are capable of solving constrained complex problems with diverse search spaces.
LoRa (Long Range) is one of the latest Low power Wide Area Network (LPWAN) technology that has increased the number of IoT applications because of its extended battery life, low data rate and large coverage area. In this paper, we have investigated and analyzed the effects of different transmission parameters of LoRa on the estimated battery life of sensors. To extend the battery life of LoRa based sensors, optimal values of non-constrained parameters such as Spreading factor (SF), Coding rate (CR) and Bandwidth (BW) has been analyzed on the basis of Mean Square Error (MSE) Function. The potency of MSE is evaluated by the means of Artificial Neural Network using neural network fitting tool in MATLAB for simulations. In comparison to current prevalent lifetime of LoRa based node i.e. 10years, the optimization insights an increase in the battery life of nodes upto 19 years. Encapsulating these benefits of LoRaWAN, this technology has been proved as the propitious methods in different and wide applications of IoT.
A novel bio-inspired evolutionary algorithm known as MoFAL is presented in this article. The proposed algorithm (MoFAL) is based on the hybrid amalgamation of two nature inspired methods based on Moth Flame Optimization and Ant Lion Optimizer algorithms. It is well known that elitism forms an important characteristic of evolutionary algorithms that allows them to maintain the best fitness(es) obtained at any stage of the optimization process. MoFal is bench-marked using a set of 23 classical benchmark functions employed to test different characteristics during its evolutionary computation process. Numerical experiments demonstrate that the solutions of the constrained optimization problems like Pressure Vessel and the Rolling Element Bearing designs found using our algorithm are highly accurate and their convergence is comparatively fast coupled with improved exploration, local optima avoidance and exploitation. The results clearly exhibit that MoFAL algorithm is capable of finding superior optimal designs for our case study problems that include diverse search spaces. Our algorithm is able to determine global solutions of constrained optimization problems more efficiently than traditional evolutionary algorithms, and also avoid the occurrence of premature phenomena during its convergence process.
Creating vegetation contents for a digital twin city entails generating dynamic 3D plant models in a large scale to represent the actual vegetation in the city. To enable high-fidelity environmental simulations and analysis applications, we model individual trees at a species level of detail. The 3D models are generated procedurally based on their botanical species profiles within the constraints of measurements and growth spaces derived from laser-scanned point cloud data. Users can conveniently define the known profile of a species by using a species profile template that we formulated based on species growth processes and patterns. Based on the given species profile and solving for the unknowns within the growth space constraints, a species model will be grown through iterations of our formulated growth rules. We show that this methodology produces structurally-representative species models with respect to their actual physical and species characteristics.
Sketch-based modelling (SBM) has undergone substantial research over the past two decades. In the early days, researchers aimed at developing techniques useful for modelling of architectural and mechanical models through sketching. With the advancement of technology used in designing visual effects for film, TV and games, the demand for highly realistic 3D character models has skyrocketed. To allow artists to create 3D character models quickly, researchers have proposed several techniques for efficient character modelling from sketched feature curves. Moreover several research groups have developed 3D shape databases to retrieve 3D models from sketched inputs. Unfortunately, the current state of the art in sketch-based organic modelling (3D character modelling) contains a lot of gaps and limitations. To bridge the gaps and improve the current sketch-based modelling techniques, this research aims to develop an approach allowing direct and interactive modelling of 3D characters from sketched feature curves, and also make use of 3D shape databases to guide the artist to create his / her desired models. The research involved finding a fusion between 3D shape retrieval, shape manipulation, and shape reconstruction / generation techniques backed by an extensive literature review, experimentation and results. The outcome of this research involved devising a novel and improved technique for sketch-based modelling, the creation of a software interface that allows the artist to quickly and easily create realistic 3D character models with comparatively less effort and learning. The proposed research work provides the tools to draw 3D shape primitives and manipulate them using simple gestures which leads to a better modelling experience than the existing state of the art SBM systems.
There is an extensive literature on dynamic algorithms for a large number of graph theoretic problems, particularly for all varieties of shortest path problems. Germane to this paper are a number fully dynamic algorithms that are known for chordal graphs. However, to the best of our knowledge no study has been done for the problem of dynamic algorithms for strongly chordal graphs. To address this gap, in this paper, we propose a semi-dynamic algorithm for edge-deletions and a semi-dynamic algorithm for edge-insertions in a strongly chordal graph, $G = (V, E)$, on $n$ vertices and $m$ edges. The query complexity of an edge-deletion is $O(d_u^2d_v^2 (n + m))$, where $d_u$ and $d_v$ are the degrees of the vertices $u$ and $v$ of the candidate edge $\{u, v\}$, while the query-complexity of an edge-insertion is $O(n^2)$.
An array with spares on four sides and the restructuring algorithm for it were proposed in [1]. However, the restructuring algorithm described in [1] is too complicated to be realized in hardware. Here, we propose a method to improve such the situation. First, the array is considered to be an (N +2) (N +2) array if four PEs are added to the four corners of the array and the spares are included. The (N+2) (N+2) array is divided into four subarrays, each of which is of size (N=2 + 1)(N=2 + 1), and the orthogonal side rotation introduced here is individually applied to each subarray. The reliabilities are given by computer simulation. They fairly increase, comparing with those in [1]. :
