Interest in research of the navigation problem for Unmanned Aerial Vehicles (UAVs) is on the rise. The aim of such a task is reaching a goal position while avoiding obstacles on the way. In this paper, we propose a different approach to Deep Reinforcement Learning (DRL) of navigation decision making process by introducing the reward function based of Artificial Potential Fields (APF). The validation of the proposed approach is performed by the comparison to the state-of-the-art approach. In terms of training performance, success rate, memory usage and the inference time, our approach, though sparser in terms of perceived information about the environment, yield better results.

This paper provides an overview of the influential parameters for the power circuit breaker condition assessment based on the vibration fingerprint. By creating the feature subsets based on the domain of computation originating from the vibration fingerprint, the features are firstly ranked by four features ranking algorithms. To confirm the ranked feature contribution to the classification performance, 11 different machine learning classifiers are trained. The training of the classifier is performed on the complete feature set where afterward the same classifiers are trained with the subset of the features ordered by the ranking algorithms. The ranking and the classifier performance yield the concept of kurtosis in the time and frequency domain as a highly promising feature for binary classification which credibly reflects the circuit breaker's mechanical condition.

This paper presents KF-RRT algorithm: a novel approach to path planning for robotic manipulators in dynamic environments. It is based on a modified RRT algorithm combined with Kalman filtering technique. RRT modification implies two aspects. The first one is related to continuous update of struc-ture/ordering within the tree to accommodate for online execution of the algorithm. The second one relies on forest-based replanning by combining connected components. On the other hand, Kalman filter is used to track/predict the motion of obstacles. Virtually augmented obstacles influence the growth of trees, which yields the improved safety margin of the resulting motion. KF-RRT is validated within a simulation study, where it is compared to comneting algorithms,

Time-aware recommender systems extend traditional recommendation methods by revealing user preferences over time or observing a specific temporal context. Among other features and advantages, they can be used to provide rating predictions based on changes in recurring time periods. Their underlying assumption is that users are similar if their behavior is similar in the same temporal context. Existing approaches usually consider separate temporal contexts and generated user profiles. In this paper, we create user profiles based on multidimensional temporal contexts and use their combined presentation in a user-based collaborative filtering method. The proposed model provides user preferences at a future point in time that matches temporal profiles. The experimental validation demonstrates that the proposed model is able to outperform the usual collaborative filtering algorithms in prediction accuracy.

Control design for trajectory tracking of multi-rotor aerial vehicles (MAVs) represents a challenging task due to the under-actuated property, highly nonlinear and cross-coupled dynamics, modeling errors, parametric uncertainties and external disturbances. This paper presents the design of the first order sliding mode control (FOSMC) algorithm for trajectory tracking of the octo-rotor unmanned aerial vehicle (UAV) in the presence of various disturbances. The highly nonlinear octo-rotor UAV dynamics is considered via the generalized framework for MAVs modeling. The stability analysis of the closed-loop system is presented using the Lyapunov based approach. The developed FOSMC exhibits finite-time convergence of the octo-rotor trajec-tories to the sliding manifold and the asymptotic stability of the equilibrium in the presence of vanishing disturbances. Simulation studies show a superior tracking performance and robustness properties of the FOSMC in comparison with the concurrent techniques for trajectory tracking of the octo-rotor UAV in the presence of internal and external disturbances.

The cloud has become an essential part of modern computing, and its popularity continues to rise with each passing day. Currently, cloud computing is faced with certain challenges that are, due to the increasing demands, becoming urgent to address. One such challenge is the problem of load balancing, which involves the proper distribution of user requests within the cloud. This paper proposes a genetic algorithm for load balancing of the received requests across cloud resources. The algorithm is based on the processing of individual requests instantly upon arrival. The conducted test simulations showed that the proposed approach has better response and processing time compared to round robin, ESCE and throttled load balancing algorithms. The algorithm outperformed an existing genetic based load balancing algorithm, DTGA, as well.

This paper presents a fine-tuned implementation of the quicksort algorithm for highly parallel multicore NVIDIA graphics processors. The described approach focuses on algorith-mic and implementation-level improvements to achieve enhanced performance. Several fine-tuning techniques are explored to identify the best combination of improvements for the quicksort algorithm on GPUs. The results show that this approach leads to a significant reduction in execution time and an improvement in algorithmic operations, such as the number of iterations of the algorithm and the number of operations performed compared to its predecessors. The experiments are conducted on an NVIDIA graphics card, taking into account several distributions of input data. The findings suggest that this fine-tuning approach can enable efficient and fast sorting on GPUs for a wide range of applications.

Implementation of credit scoring models is a demanding task and crucial for risk management. Wrong decisions can significantly affect revenue, increase costs, and can lead to bankruptcy. Together with the improvement of machine learning algorithms over time, credit models based on novel algorithms have also improved and evolved. In this work, novel deep neural architectures, Stacked LSTM, and Stacked BiLSTM combined with SMOTE oversampling technique for the imbalanced dataset were developed and analyzed. The reason for the lack of publications that utilize Stacked LSTM-based models in credit scoring lies exactly in the fact that the deep learning algorithm is tailored to predict the next value of the time series, and credit scoring is a classification problem. The challenge and novelty of this approach involved the necessary adaptation of the credit scoring dataset to suit the time sequence nature of LSTM-based models. This was particularly crucial as, in practical credit scoring datasets, instances are not correlated nor time dependent. Moreover, the application of SMOTE to the newly constructed three-dimensional array served as an additional refinement step. The results show that techniques and novel approaches used in this study improved the performance of credit score prediction.

The main focus of this study is early-stage flame detection, where the number of flame pixels in the image is very scarce. To address this challenge, a custom-made dataset was created specifically for early-stage flame detection, encom-passing challenging environmental conditions. The DeepLabv3+ architecture with ResNet-50 backbone was employed for training and weighted cross-entropy was used to effectively handle the imbalanced nature of the dataset. As a result, the model achieved a mean Intersection over Union (mIoU) value of 0.7519, demonstrating robust performance in challenging conditions. The model exhibited accurate flame pixel detection and flame shape identification in images with low flame content but high smoke levels. Additionally, the model performed well in night-time conditions, accurately identifying flame regions and shapes. An important aspect of the model's performance was its ability to correctly identify images with no flames, thereby reducing false alarms and making it suitable for UAV-based flame detection tasks.

With the advancements of Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL), it is now possible to greatly speed up the processes of predicting certain anomalies and prevent unforeseen situations and disasters. One example of such an environmental disaster is the problem of early-stage flame segmentation. It is not only important to create a model capable of pattern recognition with high accuracy but also to optimize it for real-time execution. In this paper, we demonstrate the capabilities of Deeplabv3+ for early-stage flame segmentation on a custom-made dataset with challenging conditions, and near real-time execution with the adoption of the Open VINO toolkit. Acceleration of the inference process in the range of 70.46% to 93.46% is achieved, while the speed of the inference process when using the GPU with FP16 precision is increased by almost 2 times when compared to FP32 precision. The impact of our findings is significant, as early-stage flame segmentation is a critical component of disaster prevention in environmental settings. Our results demonstrate the potential of using the OpenVINO toolkit for the acceleration of the inference process.

The prediction of the dynamics of High-Speed Craft (HSC) with prismatic hulls is commonly performed by designers using semi-empirical formulations based on Savitsky’s classic method. However, the accuracy of this prediction decreases with the presence of warp, when the deadrise of the hull change along its length, which is typical for small passenger ferries, even when considering the effective deadrise and trim angle concept proposed by Savitsky in 2012. The present work assessed the dynamics of three planing warped hulls and one prismatic monohull developed by the University of Naples, using a morphing grid approach implemented in OpenFOAM to capture the motion of the vessel. Numerical results on resistance, wetted area, dynamic trim angle, wall shear stress, and pressure distribution were compared with the method proposed by Savitsky, and previously published results where possible. Results suggested that it is possible to improve Savitsky prediction by changing the location where the equivalent deadrise angle is evaluated. This single modification will allow to extend the application of Savitsky method for a wider range of warp rates.

To maximize the impact of precision medicine approaches, it is critical to accurately identify genetic variants in cancer-associated genes with functional consequences. Yet, our knowledge of rare variants conferring clinically relevant phenotypes and the mechanisms through which they act remains highly limited. A tumor suppressor gene exemplifying the challenge of variant interpretation is VHL. VHL encodes an E3 ubiquitin ligase that regulates the cellular response to hypoxia. Germline pathogenic variants in VHL predispose patients to tumors including clear cell renal cell carcinoma (ccRCC) and pheochromocytoma, and somatic VHL mutations are frequently observed in sporadic renal cancer. Here, we optimize and apply Saturation Genome Editing (SGE) to assay nearly all possible single nucleotide variants (SNVs) across VHL’s coding sequence. To delineate mechanisms, we quantify mRNA dosage effects over time and compare effects in isogenic cell lines. Function scores for 2,268 VHL SNVs identify a core set of pathogenic alleles driving ccRCC with perfect accuracy, inform differential risk across tumor types, and reveal novel mechanisms by which variants impact function. These results have immediate utility for classifying VHL variants encountered in both germline testing and tumor profiling and illustrate how precise functional measurements can resolve pleiotropic and dosage-dependent genotype-phenotype relationships across complete genes.

Road infrastructure management is an extremely important task of traffic engineering. For the purpose of efficient management, it is necessary to determine the efficiency of the traffic flow through PAE 85%, AADT and other exploitation parameters on the one hand, and the number of different types of traffic accidents on the other. In this paper, a novel TrIT2F (trapezoidal interval type-2 fuzzy) PIPRECIA (pivot pairwise relative criteria importance assessment)-TrIT2F MARCOS (measurement of alternatives and ranking according to compromise solution) was developed in order to, in a defined set of 14 road segments, identify the most efficient one for data related to light goods vehicles. Through this the aims and contributions of the study can be manifested. The evaluation was carried out on the basis of seven criteria with weights obtained using the TrIT2F PIPRECIA, while the final results were presented through the TrIT2F MARCOS method. To average part of the input data, the Dombi and Bonferroni operators have been applied. The final results of the applied TrIT2F PIPRECIA-TrIT2F MARCOS model show the following ranking of road segments, according to which Vrhovi–Šešlije M-I-103 with a gradient of −1.00 represents the best solution: A5 > A8 > A2 > A1 > A4 > A3 > A6 > A12 > A13 = A14 > A11 > A7 > A9 > A10. In addition, the validation of the obtained results was conducted by changing the values of the four most important criteria and changing the size of the decision matrix. Tests have shown great stability of the developed TrIT2F PIPRECIA-TrIT2F MARCOS model.

AIM To critically evaluate the reporting quality of a random sample of animal studies within the field of endodontics against the Preferred Reporting Items for Animal Studies in Endodontics (PRIASE) 2021 checklist and to investigate the association between the quality of reporting and several characteristics of the selected studies. METHODOLOGY Fifty animal studies related to endodontics were randomly selected from the PubMed database with publication dates from January 2017 to December 2021. For each study, a score of '1' was given when the item of the PRIASE 2021 checklist was fully reported, whereas a score of '0' was given when an item was not reported; when the item was inadequately or partially reported, a score of '0.5' was given. Based on the overall scores allocated to each manuscript, they were allocated into three categories of reporting quality: low, moderate, and high. Associations between study characteristics and reporting quality scores were also analysed. Descriptive statistics, and Fisher's exact tests were used to describe the data and determine associations. The probability value of 0.05 was selected as the level of statistical significance. RESULTS Based on the overall scores, four (8%) and 46 (92%) of the animal studies evaluated were categorised as 'High' and 'Moderate' reporting quality, respectively. A number of items were adequately reported in all studies related to background (Item 4a), relevance of methods/results (7a) and interpretation of images (11e), whereas only one item related to changes in protocol (6d) was not reported in any. No associations were confirmed between reporting quality scores and number of authors, origin of the corresponding author, journal of publication (endodontic specialty versus non- specialty), impact factor or year of publication. CONCLUSIONS Animal studies published in the specialty of endodontics were mostly of 'moderate' quality in terms of the quality of reporting. Adherence to the PRIASE 2021 guidelines will enhance the reporting of animal studies in the expectation that all future publications will be high-quality.

Leveraging second-order information at the scale of deep networks is one of the main lines of approach for improving the performance of current optimizers for deep learning. Yet, existing approaches for accurate full-matrix preconditioning, such as Full-Matrix Adagrad (GGT) or Matrix-Free Approximate Curvature (M-FAC) suffer from massive storage costs when applied even to medium-scale models, as they must store a sliding window of gradients, whose memory requirements are multiplicative in the model dimension. In this paper, we address this issue via an efficient and simple-to-implement error-feedback technique that can be applied to compress preconditioners by up to two orders of magnitude in practice, without loss of convergence. Specifically, our approach compresses the gradient information via sparsification or low-rank compression \emph{before} it is fed into the preconditioner, feeding the compression error back into future iterations. Extensive experiments on deep neural networks for vision show that this approach can compress full-matrix preconditioners by up to two orders of magnitude without impact on accuracy, effectively removing the memory overhead of full-matrix preconditioning for implementations of full-matrix Adagrad (GGT) and natural gradient (M-FAC). Our code is available at https://github.com/IST-DASLab/EFCP.