Explore the vast range of titles available.

Deep neural networks (DNNs) have achieved outstanding results in domains such as image processing, computer vision, natural language processing and bioinformatics. In recent years, many methods have been proposed that can provide a visual explanation of decision made by such classifiers. Saliency maps are probably the most popular. However, it is still unclear how to properly interpret saliency maps for a given image and which techniques perform most accurately. This paper presents a methodology to practically evaluate the real effectiveness of saliency map generation methods. We used three state-of-the-art network architectures along with specially prepared benchmark datasets, and we proposed a novel metric to provide a quantitative comparison of the methods. The comparison identified the most reliable techniques and the solutions which usually failed in our tests.

This paper considers the black-box approximation problem where the goal is to create a regression model using only empirical data without incorporating knowledge about the character of nonlinearity of the approximated function. This paper reports on ongoing work on a nonlinear regression methodology called IBHM which builds a model being a combination of weighted nonlinear components. The construction process is iterative and is based on correlation analysis. Due to its iterative nature, the methodology does not require a priori assumptions about the final model structure which greatly simplifies its usage. Correlation based learning becomes ineffective when the dynamics of the approximated function is too high. In this paper we introduce weighted correlation coefficients into the learning process. These coefficients work as a kind of a local filter and help overcome the problem. Proof of concept experiments are discussed to show

A soft computing system used to optimize deep drilling operations under high-speed conditions in the manufacture of steel components is presented. The input data includes cutting parameters and axial cutting force obtained from the power consumption of the feed motor of the milling centres. Two different coolant strategies are tested: traditional working fluid and Minimum Quantity Lubrication (MQL). The model is constructed in three phases. First, a new strategy is proposed to evaluate and complete the set of available measurements. The primary objective of this phase is to decide whether further drilling experiments are required to develop an accurate roughness prediction model. An important aspect of the proposed strategy is the imputation of missing data, which is used to fully exploit both complete and incomplete measurements. The proposed imputation algorithm is based on a genetic algorithm and aims to improve prediction accuracy. In the second phase, a bag of multilayer perceptrons is used to model the impact of deep drilling settings on borehole roughness. Finally, this model is supplied with the borehole dimensions, coolant option and expected axial force to develop a 3D surface showing the expected borehole roughness as a function of drilling process settings. This plot is the necessary output of the model for its use under real workshop conditions. The proposed system is capable of approximating the optimal model used to control deep drilling tasks on steel components for industrial use.

Session-based recommenders, used for making predictions out of users' uninterrupted sequences of actions, are attractive for many applications. Here, for this task we propose using metric learning, where a common embedding space for sessions and items is created, and distance measures dissimilarity between the provided sequence of users' events and the next action. We discuss and compare metric learning approaches to commonly used learning-to-rank methods, where some synergies exist. We propose a simple architecture for problem analysis and demonstrate that neither extensively big nor deep architectures are necessary in order to outperform existing methods. The experimental results against strong baselines on four datasets are provided with an ablation study.

Worldwide, there is a significant increase in the use of unmanned aerial vehicles (UAVs) by emergency services. They offer a lot of possibilities during rescue operations. Such a wide application for various purposes and environments causes many threats related to their use. To minimize the risks associated with conducting air operations with UAVs, the application of the SORA (Specific Operations Risk Assessment) methodology will be important. Due to its level of detail, it is a methodology adapted to civilian use. In this article, the authors’ team will try to develop guidelines and directions for adapting SORA to the requirements of the operational work of emergency services. Thus, the following article aims to present the most important risks related to conducting operations with the use of UAVs by first responders (FRs), and to show the sample risk analysis performed for this type of operation on the example of the ASSISTANCE project. The paper describes, on the one hand, possibilities offered by UAVs in crisis or disaster management and step-by-step Specific Operations Risk Assessment (SORA), and on the other hand, presents possible threats, consequences and methods of their mitigation during FR missions.

Geothermal energy is a crucial component contributing to the development of local thermal energy systems as a carbon-neutral and reliable energy source. Insights into its availability derive from knowledge of geology, hydrogeology and the thermal regime of the subsurface. This expertise helps to locate and monitor geothermal installations as well as observe diverse aspects of natural and man-made thermal effects. Temperature measurements were performed in hydrogeological boreholes in south-western Poland using two methods, i.e., manual temperature logging and optical fibre distributed temperature sensing (OF DTS). It was assumed the water column in each borehole was under thermodynamic equilibrium with the local geothermal gradient of the subsurface, meaning rocks and aquifers. Most of the acquired results show typical patterns, with the upper part of the log depending on altitude, weather and climate as well as on seasonal temperature changes. For deeper parts, the temperature normally increases depending on the local geothermal gradient. The temperature logs for some boreholes located in urban agglomerations showed anthropogenic influence caused by the presence of infrastructure, the urban heat island effect, post-mining activities, etc. The presented research methods are suitable for applications connected with studies crucial to selecting the locations of geothermal installations and to optimize their technical parameters. The observations also help to identify zones of intensified groundwater flow, groundwater inrush into wells, fractured and fissured zones and many others.

Investment and construction plans, architectural and construction decisions, and spatial and technology-related decisions made at the early stages of a project have a significant impact on meeting the investment goals and customer expectations. Decision making is a very time-consuming and complicated process (due to the complexity of construction processes). The whole difficulty comes to specifying the appropriate criteria for assessing the given activities, providing answers to the questions of the decision-making bodies. A set of appropriate criteria and mathematical tools (such as computer algorithms with multi-criteria analysis) can significantly improve and accelerate the decision-making process. This article combines ESORD (an IT tool that allows you to compare different types of solutions based on mathematical calculations) with the Monte Carlo method. The developed approach can help the investor to optimize their cash-flow schedule. The original method enables the client to select a construction project variant characterized by the best economical and sustainable parameters, while taking into account customers’ demands.

We present a single-turn agent for graphical user interface (GUI) interaction tasks, using Vision-Language Model Florence-2-Base. The agent's primary task is identifying the screen coordinates of the UI element corresponding to the user's command. It demonstrates strong performance on Screenspot and OmniAct, while maintaining a compact size of 0.27B parameters and minimal latency. Relevant improvement comes from multi-task training and MLLM-based data augmentation. Manually annotated corpora are scarce, but we show that MLLM augmentation might produce better results. On Screenspot and OmniAct, our model outperforms both GUI-specific models (e.g., SeeClick) and MLLMs (e.g., GPT-4V).