Modern foundation models provide highly expressive visual representations, yet adapting them to high-resolution medical imaging remains challenging due to limited annotations and weak supervision. Mammography, in particular, is characterized by large images, variable multi-view studies and predominantly breast-level labels, making end-to-end fine-tuning computationally expensive and often impractical. We propose Multiple Instance Learning on Precomputed Features (MIL-PF), a scalable framework that combines frozen foundation encoders with a lightweight MIL head for mammography classification. By precomputing the semantic representations and training only a small task-specific aggregation module (40k parameters), the method enables efficient experimentation and adaptation without retraining large backbones. The architecture explicitly models the global tissue context and the sparse local lesion signals through attention-based aggregation. MIL-PF achieves state-of-the-art classification performance at clinical scale while substantially reducing training complexity. We release the code for full reproducibility.

Mammography is the gold standard for the detection and diagnosis of breast cancer. This procedure can be significantly enhanced with Artificial Intelligence (AI)-based software, which assists radiologists in identifying abnormalities. However, training AI systems requires large and diverse datasets, which are often difficult to obtain due to privacy and ethical constraints. To address this issue, the paper introduces MAMmography ensemBle mOdel (MAMBO), a novel patch-based diffusion approach designed to generate full-resolution mammograms. Diffusion models have shown breakthrough results in realistic image generation, yet few studies have focused on mammograms, and none have successfully generated high-resolution outputs required to capture fine-grained features of small lesions. To achieve this, MAMBO integrates separate diffusion models to capture both local and global (image-level) contexts. The contextual information is then fed into the final model, significantly aiding the noise removal process. This design enables MAMBO to generate highly realistic mammograms of up to 3840x3840 pixels. Importantly, this approach can be used to enhance the training of classification models and extended to anomaly segmentation. Experiments, both numerical and radiologist validation, assess MAMBO's capabilities in image generation, super-resolution, and anomaly segmentation, highlighting its potential to enhance mammography analysis for more accurate diagnoses and earlier lesion detection. The source code used in this study is publicly available at: https://github.com/iai-rs/mambo.

The surge in large language models (LLMs) has greatly advanced natural language processing. However, their development is often hampered by the limited availability and quality of training datasets, particularly for underrepresented languages. Our study aims to address this by enriching Serbian language resources, a crucial step in enhancing LLMs. A key aspect of our methodology is the use of 12 specialized metrics, which allow us to rigorously evaluate the dataset's integrity and reveal linguistic trends in the Serbian digital context. Additionally, we employ a modified Interquartile Range (IQR) method for more stringent outlier detection, aiding in the identification of linguistic anomalies in Serbian text.

Monitoring soil organic carbon (SOC) typically assumes conducting a labor-intensive soil sampling campaign, followed by laboratory testing, which is both expensive and impractical for generating useful, spatially continuous data products. The present study leverages the power of machine learning (ML) and, in particular, deep neural networks (DNNs) for segmentation, as well as satellite imagery, to estimate the SOC remotely. We propose a new two-stage pipeline for remote SOC estimation, which relies on using a DNN trained to classify land cover to perform feature extraction, while the SOC estimation is performed by a different ML model. The first stage is an image segmentation DNN with the U-Net architecture, which is trained to estimate the land cover for an observed geographical region, based on multi-spectral images taken by the Sentinel-2 satellite constellation. This estimator is subsequently used to extract the latent feature vector for each of the output pixels, by rolling back from the output (dense) layer of the U-Net and accessing the last available convolutional layer of the same dimension as our desired output. The second stage is trained on a set of feature vectors extracted at the coordinates for which manual SOC measurements exist. We tested a variety of ML models and report on their performance. Using the best extremely randomized trees model, we generated a spatially continuous map of SOC estimations for the region of Tuscany, in Italy, with a resolution of 10 m, to share with the researchers as a means of validating the results and to demonstrate the efficiency of the proposed approach, which can can easily be scaled to create a global continuous SOC map.

The COVID-19 pandemic's rapid growth has made it crucial to develop reliable and efficient diagnostic methods. In this study, we incorporate deep features and handcrafted features to provide a unique method for COVID-19 identification using chest X-rays. In order to extract high-level features from the chest X-ray pictures, we first use a convolutional neural network (CNN) that has already been trained to take advantage of deep learning. The discriminative information regarding COVID-19 infection is captured by the obtained deep features. In addition to the deep features, we also use manually created features that are meant to capture the unique features of COVID-19 in chest X-rays. Based on earlier study findings and domain understanding, these characteristics were manually constructed. They consist of statistical measures, shape-based characteristics, and texture descriptors. Comparing the performance of the classification with the standalone applications of convolutional and handcrafted features, we find that combining the features in our innovative framework enhances performance.

Mammography is the leading methodology used to diagnose breast cancer. Effective, cheap and reliable, the mammography can be used to screen large populations, if the imagery produced can be analysed efficiently. State-of-the-art generative artificial intelligence approaches can be used to create tools able to aid in this task. Here we present a study focused on the emerging research topic of the application of generative diffusion models to the task of anomaly detection and we apply if for detecting anomalies on mammograms. Diffusion models exhibit promising results in making pixel-level predictions with image level annotations, but no such application has been published so far regarding mammography. We have, therefore, developed a novel approach utilizing U-net backbone that is able to generate mammograms with Frechet Inception Distance (FID) of 14.62. We showed its ability to perform anomaly detection with Intersection over Union (IoU) of 0.195 which demonstrates the viability of our approach for early-stage research.

This paper explores the potential of a multidisciplinary approach to testing and aligning artificial intelligence (AI), specifically focusing on large language models (LLMs). Due to the rapid development and wide application of LLMs, challenges such as ethical alignment, controllability, and predictability of these models emerged as global risks. This study investigates an innovative simulation-based multi-agent system within a virtual reality framework that replicates the real-world environment. The framework is populated by automated 'digital citizens,' simulating complex social structures and interactions to examine and optimize AI. Application of various theories from the fields of sociology, social psychology, computer science, physics, biology, and economics demonstrates the possibility of a more human-aligned and socially responsible AI. The purpose of such a digital environment is to provide a dynamic platform where advanced AI agents can interact and make independent decisions, thereby mimicking realistic scenarios. The actors in this digital city, operated by the LLMs, serve as the primary agents, exhibiting high degrees of autonomy. While this approach shows immense potential, there are notable challenges and limitations, most significantly the unpredictable nature of real-world social dynamics. This research endeavors to contribute to the development and refinement of AI, emphasizing the integration of social, ethical, and theoretical dimensions for future research.

This paper provides relevant outcomes regarding the design and development of an IoT learning system, aimed to stimulate learning by doing while offering IoT deployment in medical science and related practices toward health monitoring and diagnostics. The system is conceptualized as a through-hole module integration board that enables data acquisition regarding body temperature, heart rate, and the saturation of oxygen in the blood, accompanied by psychrometric parameters of the environment in which the measurement is performed. The paper provides an overview in terms of favoring the development of digital technologies curriculums due to the importance and influences it might have on future experts. Last but not least, technical specifications, schematics, and overall system appearance are enclosed in detail. Finally, possible applications in process engineering and thermo-energetics are discussed as well.

Large-scale infrastructure, such as China–Europe Railway Express (CER-Express), which connects countries and regions across Asia and Europe, has a potentially profound effect on land use, as evidenced by changes in land cover along the railway. To ensure sustainable development of such infrastructure and appropriate land administration, effective ways to monitor and assess its impact need to be developed. Remote sensing based on publicly available satellite imagery represents an obvious choice. In the study presented here, we employ a state-of-the-art deep-learning-based approach to automatically detect different types of land cover based on multispectral Sentinel-2 imagery. We then use these data to conduct and present a study of the changes in land use in two geopolitically diverse regions of interest (in Serbia and China and with and without CER-Express infrastructure) for the period of the last three years. Our results show that the standard image-patch-based land cover classification approaches suffer a significant drop in performance in our target scenario in which each pixel needs to be assigned a cove class, but still, validate the applicability of the proposed approach as a remote sensing tool to support the sustainable development of large infrastructure. We discuss the technical limitations of the proposed approach in detail and potential ways in which it can be improved.

Abstract This work describes an approach to calculate pedological parameter maps using hyperspectral remote sensing and soil sensors. These maps serve as information basis for automated and precise agricultural treatments by tractors and field robots. Soil samples are recorded by a handheld hyperspectral sensor and analyzed in the laboratory for pedological parameters. The transfer of the correlation between these two data sets to aerial hyperspectral images leads to 2D-parameter maps of the soil surface. Additionally, rod-like soil sensors provide local 3D-information of pedological parameters under the soil surface. The goal is to combine the area-covering 2D-parameter maps with the local 3D-information to extrapolate large-scale 3D-parameter maps using AI approaches.

The China-Europe Railway Express (CER-Express), not only promoted the cooperation between countries and regions across Asia and Europe but also gave rise to remarkable changes in landcover and had a profound effect on the natural environment along the railway in recent years. Effective ways to monitor and assess ecological changes are urgently needed to ensure sustainable development of CER-Express. There are very few existing environmental monitoring studies focusing on the area along the CERExpress. In this paper, we present a study of environmental degradation, which occurred during the construction and operation of CER-Express from 2010 to 2018, based on a comprehensive evaluation index (CEI), which takes three environmental indicators into account and provides a timely and reliable evaluation of environmental changes at large scales. In addition, the environment conditions of the regions and countries along the CER-Express have been quantified and comparatively studied at different scales over different periods, using histograms of mean CEI values. Furthermore, specific causes of environmental degradation in the rail-intensive countries and small-area countries along the railway are discussed. Our results show that the environmental degradation can be detected in most of the rail-intensive countries, such as Germany, Poland, Austria, and Czech Republic along the railway. Therefore, to ensure sustainability of the CER-Express, environmental protection along the railway should be paid more attention to and a reasonable arrangement for the exploitation of CER-Express devised.

In this paper, we present an initial study of possibilities of applying Artificial Intelligence (AI) and computer vision-based approaches aimed to improve and alleviate the process of conducting knowledge assessment over the Microsoft Teams platform. We did that by developing a deep-neural-network-based system which is able to locate faces and predict emotions based on students' facial expressions. The system was evaluated on videos recorded during an online assessment of the ability of students to train and deploy deep learning solutions using Python and TensorFlow. We present results of this evaluation and show that, although the accuracy of our algorithm is limited at frame level, as we optimized for computational performance, it provides sufficient information to identify key changes on students' behavior, which should be brought to the teachers' attention.

The results presented in this contribution demonstrate the value of climate services for the planned construction of the new Wastewater Treatment Plant (WWTP) in Novi Sad, Serbia. In this case, climate services provided added value for the decision-making processes, in terms of enhanced effectiveness, optimized technological opportunities and minimized risks and by serving as the means of involving and better-informing end-users and stakeholders. The specific goal of the research was to improve climate change resilience of the WWTP and to facilitate better overall hygienic conditions in Novi Sad and to safeguard the potable water resources and the quality of the environment in the areas located downstream and under the influence of the Danube River. In order to achieve it, preliminary activities were oriented on analyzing the current climate and hydrological conditions, engaging the relevant data providers, stakeholders and policy makers and evaluating what relevant local data would be useful for the study. The data collected was applied in the testing and for improving the Future Danube Multi-hazard, Multi-risk Model (FDM), a catastrophe model implemented in the OASIS Loss Modeling Framework (Oasis-LMF). The FDM is implemented for the entire Danube Basin. High-resolution components for pluvial flood risks were further implemented to the city of Novi Sad, Serbia, after successful testing in the Budapest region. Observations and model results were used in a climate change impact assessment with the purpose of identifying adaptation options, appraisal of adaptation options and integration of an adaptation action plan into the Feasibility Study of the WWTP construction. The results of the pluvial flood model for Novi Sad clearly suggested that it is important to consider pluvial flood risks and that protective measures have to be considered as part of the WWTP construction, both under current and future climate conditions. Moreover, novel estimates of drainage water intensities during heavy rains would advise the design of the simultaneously planned pumping station on the banks of the Danube. Combined, this clearly demonstrates the added value of the climate services and risk information delivered by the FDM also beyond the insurance sector, as well as its potential to support adaptation decision making with respect to infrastructural investments in Novi Sad.

Minimizing last mile delivery costs is of paramount importance for all shipping companies that strive to stay competitive on the market. A potential solution to the problem is the use of crowdsourcing - a model where individuals voluntarily take on a task proposed by another entity (e.g. a company). In this paper, we present the results of a comparison of performance for three types of crowdsourced delivery fleets likely to be used in an urban setting. The fleets differ in the mode of transport the couriers use: bicycles, cars or both. The performance is quantified by the total number of deliveries made and the on-time delivery rates. Experimental results were obtained through a simulation that closely resembles real-world traffic conditions in a city with developed cycling infrastructure and takes into account the variations in the speed of couriers. The research shows that bicycle-based crowdsourced fleets outperform other kinds of fleets under simulated conditions. This makes them a faster, more environmentally-friendly and potentially cheaper alternative to traditional fleets that rely on cars.

Medical imaging is invaluable when it comes to gaining insight into the human body. As is well known, medical images need to deal with artefacts. This article presents a modern procedure for metal artifact reduction in computed tomography, which relies on additional information extracted from corresponding magnetic resonance images. We conducted a simulation study so as to compare the resulting images with those corrected, using the baseline linear interpolation method. The outcome indicates that the proposed method incomparably outperforms the baseline and reduces metal artefacts, improving the quality of images, which can be later used in a clinical setting.

This chapter presents an overview of the published research focused on the application of visual attention and saliency models to the problem of image and video quality assessment. Determining the perceptual quality of multimedia content is crucial for achieving quality-of-experience-driven multimedia services. The problem has been gaining significance in the wake of the recent explosion of visual and multimedia applications.Attention and saliency models have the potential to improve the performance of state-of-the-art quality assessment algorithms significantly and are generating increased interest within the research community.

The latest generation of convolutional neural networks (CNNs) has achieved impressive results in the field of image classification. This paper is concerned with a new approach to the development of plant disease recognition model, based on leaf image classification, by the use of deep convolutional networks. Novel way of training and the methodology used facilitate a quick and easy system implementation in practice. The developed model is able to recognize 13 different types of plant diseases out of healthy leaves, with the ability to distinguish plant leaves from their surroundings. According to our knowledge, this method for plant disease recognition has been proposed for the first time. All essential steps required for implementing this disease recognition model are fully described throughout the paper, starting from gathering images in order to create a database, assessed by agricultural experts. Caffe, a deep learning framework developed by Berkley Vision and Learning Centre, was used to perform the deep CNN training. The experimental results on the developed model achieved precision between 91% and 98%, for separate class tests, on average 96.3%.

The burden of chronic disease and associated disability present a major threat to financial sustainability of healthcare delivery systems. The need for cost-effective early diagnosis and disease prevention is evident driving the development of personalized home health solutions. The proposed solution presents an easy to use ECG monitoring system. The core hardware component is a biosensor dongle with sensing probes at one end, and micro USB interface at the other end, offering reliable and unobtrusive sensing, preprocessing and storage. An additional component is a smart phone, providing both the biosensor's power supply and an intuitive user application for the real-time data reading. The system usage is simplified, with innovative solutions offering plug and play functionality avoiding additional driver installation. Personalized needs could be met with different sensor combinations enabling adequate monitoring in chronic disease, during physical activity and in the rehabilitation process.

An increasing amount of geo-referenced mobile phone data enables the identification of behavioral patterns, habits and movements of people. With this data, we can extract the knowledge potentially useful for many applications including the one tackled in this study - understanding spatial variation of epidemics. We explored the datasets collected by a cell phone service provider and linked them to spatial HIV prevalence rates estimated from publicly available surveys. For that purpose, 224 features were extracted from mobility and connectivity traces and related to the level of HIV epidemic in 50 Ivory Coast departments. By means of regression models, we evaluated predictive ability of extracted features. Several models predicted HIV prevalence that are highly correlated (>0.7) with actual values. Through contribution analysis we identified key elements that correlate with the rate of infections and could serve as a proxy for epidemic monitoring. Our findings indicate that night connectivity and activity, spatial area covered by users and overall migrations are strongly linked to HIV. By visualizing the communication and mobility flows, we strived to explain the spatial structure of epidemics. We discovered that strong ties and hubs in communication and mobility align with HIV hot spots.

We address the problem of automatic extraction of foreground objects from videos. The goal is to provide a method for unsupervised collection of samples which can be further used for object detection training without any human intervention. We use the well known Selective Search approach to produce an initial still-image based segmentation of the video frames. This initial set of proposals is pruned and temporally extended using optical flow and transductive learning. Specifically, we propose to use Dense Trajectories in order to robustly match and track candidate boxes over different frames. The obtained box tracks are used to collect samples for unsupervised training of track-specific detectors. Finally, the detectors are run on the videos to extract the final tubes. The combination of appearance-based static "objectness" (Selective Search), motion information (Dense Trajectories) and transductive learning (detectors are forced to "overfit" on the unsupervised data used for training) makes the proposed approach extremely robust. We outperform state-of-the-art systems by a large margin on common benchmarks used for tube proposal evaluation.