Detection of COVID-19 and its pulmonary stage using Bayesian hyperparameter optimization and deep feature selection methods

Since the first case of COVID-19 was reported in December 2019, many studies have been carried out on artificial intelligence for the rapid diagnosis of the disease to support health services. Therefore, in this study, we present a powerful approach to detect COVID-19 and COVID-19 findings from computed tomography images using pre-trained models using two different datasets. COVID-19, influenza A (H1N1) pneumonia, bacterial pneumonia and healthy lung image classes were used in the first dataset. Consolidation, crazy-paving pattern, ground-glass opacity, ground-glass opacity and consolidation, ground-glass opacity and nodule classes were used in the second dataset. The study consists of four steps. In the first two steps, distinctive features were extracted from the final layers of the pre-trained ShuffleNet, GoogLeNet and MobileNetV2 models trained with the datasets. In the next steps, the most relevant features were selected from the models using the Sine–Cosine optimization algorithm. Then, the hyperparameters of the Support Vector Machines were optimized with the Bayesian optimization algorithm and used to reclassify the feature subset that achieved the highest accuracy in the third step. The overall accuracy obtained for the first and second datasets is 99.46% and 99.82%, respectively. Finally, the performance of the results visualized with Occlusion Sensitivity Maps was compared with Gradient-weighted class activation mapping. The approach proposed in this paper outperformed other methods in detecting COVID-19 from multiclass viral pneumonia. Moreover, detecting the stages of COVID-19 in the lungs was an innovative and successful approach.     

Evolutionary design of neural network architectures: a review of three decades of research

We present a comprehensive review of the evolutionary design of neural network architectures. This work is motivated by the fact that the success of an Artificial Neural Network (ANN) highly depends on its architecture and among many approaches Evolutionary Computation, which is a set of global-search methods inspired by biological evolution has been proved to be an efficient approach for optimizing neural network structures. Initial attempts for automating architecture design by applying evolutionary approaches start in the late 1980s and have attracted significant interest until today. In this context, we examined the historical progress and analyzed all relevant scientific papers with a special emphasis on how evolutionary computation techniques were adopted and various encoding strategies proposed. We summarized key aspects of methodology, discussed common challenges, and investigated the works in chronological order by dividing the entire timeframe into three periods. The first period covers early works focusing on the optimization of simple ANN architectures with a variety of solutions proposed on chromosome representation. In the second period, the rise of more powerful methods and hybrid approaches were surveyed. In parallel with the recent advances, the last period covers the Deep Learning Era, in which research direction is shifted towards configuring advanced models of deep neural networks. Finally, we propose open problems for future research in the field of neural architecture search and provide insights for fully automated machine learning. Our aim is to provide a complete reference of works in this subject and guide researchers towards promising directions.

     

Silver Nanoflower Decorated Graphene Oxide Sponges for Highly Sensitive Variable Stiffness Stress Sensors

Soft conductive materials should enable large deformation while keeping high electrical conductivity and elasticity. The graphene oxide (GO)‐based sponge is a potential candidate to endow large deformation. However, it typically exhibits low conductivity and elasticity. Here, the highly conductive and elastic sponge composed of GO, flower‐shaped silver nanoparticles (AgNFs), and polyimide (GO‐AgNF‐PI sponge) are demonstrated. The average pore size and porosity are 114 µm and 94.7%, respectively. Ag NFs have thin petals (8–20 nm) protruding out of the surface of a spherical bud (300–350 nm) significantly enhancing the specific surface area (2.83 m² g^?1). The electrical conductivity (0.306 S m^?1 at 0% strain) of the GO‐AgNF‐PI sponge is increased by more than an order of magnitude with the addition of Ag NFs. A nearly perfect elasticity is obtained over a wide compressive strain range (0–90%). The strain‐dependent, nonlinear variation of Young's modulus of the sponge provides a unique opportunity as a variable stiffness stress sensor that operates over a wide stress range (0–10 kPa) with a high maximum sensitivity (0.572 kPa^?1). It allows grasping of a soft rose and a hard bottle, with the minimal object deformation, when attached on the finger of a robot gripper.     

Fast computation of the prime implicants by exact direct-cover algorithm based on the new partial ordering operation rule

In this study, a novel OFF-set based direct-cover Exact Minimization Algorithm (EMA) is proposed for single-output Boolean functions represented in a sum-of-products form. To obtain the complete set of prime implicants covering the given Target Minterm (ON-minterm), the proposed method uses OFF-cubes (OFF-minterms) expanded by this Target Minterm. The amount of temporary results produced by this method does not exceed the size of the OFF-set. In order to achieve the goal of this study, which is to make faster computations, logic operations were used instead of the standard operations. Expansion OFF-cubes, commutative absorption operations and intersection operations are realized by logic operations for fast computation. The proposed minimization method is tested on several classes of benchmarks and then compared with the ESPRESSO algorithm. The results show that the proposed algorithm obtains more accurate and faster results than ESPRESSO does.     

Prediction of deflection of reinforced concrete shear walls

Reinforced concrete shear walls are used in tall buildings for efficiently resisting lateral loads. Due to the low tensile strength of concrete, reinforced concrete shear walls tend to behave in a nonlinear manner with a significant reduction in stiffness, even under service loads. To accurately assess the lateral deflection of shear walls, the prediction of flexural and shear stiffness of these members after cracking becomes important. In the present study, an iterative analytical procedure which considers the cracking in the reinforced concrete shear walls has been presented. The effect of concrete cracking on the stiffness and deflection of shear walls have also been investigated by the developed computer program based on the iterative procedure. In the program, the variation of the flexural stiffness of a cracked member has been evaluated by ACI and probability-based effective stiffness model. In the analysis, shear deformation which can be large and significant after development of cracks is also taken into account and the variation of shear stiffness in the cracked regions of members has been considered by using effective shear stiffness model available in the literature. Verification of the proposed procedure has been confirmed from series of reinforced concrete shear wall tests available in the literature. Comparison between the analytical and experimental results shows that the proposed analytical procedure can provide an accurate and efficient prediction of both the deflection and flexural stiffness reduction of shear walls with different height to width ratio and vertical load. The results of the analytical procedure also indicate that the percentage of shear deflection in the total deflection increases with decreasing height to width ratio of the shear wall.     

Multi‐level reranking approach for bug localization

Bug fixing has a key role in software quality evaluation. Bug fixing starts with the bug localization step, in which developers use textual bug information to find location of source codes which have the bug. Bug localization is a tedious and time consuming process. Information retrieval requires understanding the programme's goal, coding structure, programming logic and the relevant attributes of bug. Information retrieval (IR) based bug localization is a retrieval task, where bug reports and source files represent the queries and documents, respectively. In this paper, we propose BugCatcher, a newly developed bug localization method based on multi‐level re‐ranking IR technique. We evaluate BugCatcher on three open source projects with approximately 3400 bugs. Our experiments show that multi‐level reranking approach to bug localization is promising. Retrieval performance and accuracy of BugCatcher are better than current bug localization tools, and BugCatcher has the best Top N, Mean Average Precision (MAP) and Mean Reciprocal Rank (MRR) values for all datasets.     

Mobile multi-view object image search

High user interaction capability of mobile devices can help improve the accuracy of mobile visual search systems. At query time, it is possible to capture multiple views of an object from different viewing angles and at different scales with the mobile device camera to obtain richer information about the object compared to a single view and hence return more accurate results. Motivated by this, we propose a new multi-view visual query model on multi-view object image databases for mobile visual search. Multi-view images of objects acquired by the mobile clients are processed and local features are sent to a server, which combines the query image representations with early/late fusion methods and returns the query results. We performed a comprehensive analysis of early and late fusion approaches using various similarity functions, on an existing single view and a new multi-view object image database. The experimental results show that multi-view search provides significantly better retrieval accuracy compared to traditional single view search.     

Nodal Superconvergence of SDFEM for Singularly Perturbed Problems

In this paper, we analyze the streamline diffusion finite element method for one dimensional singularly perturbed convection-diffusion-reaction problems. Local error estimates on a subdomain where the solution is smooth are established. We prove that for a special group of exact solutions, the nodal error converges at a superconvergence rate of order (ln ε ⁻¹/N)^2k (or (ln N/N)^2k ) on a Shishkin mesh. Here ε is the singular perturbation parameter and 2N denotes the number of mesh elements. Numerical results illustrating the sharpness of our theoretical findings are displayed.