The high computational costs of large language models (LLMs) have led to a flurry of research on LLM compression, via methods such as quantization, sparsification, or structured pruning. A new frontier in this area is given by \emph{dynamic, non-uniform} compression methods, which adjust the compression levels (e.g., sparsity) per-block or even per-layer in order to minimize accuracy loss, while guaranteeing a global compression threshold. Yet, current methods rely on heuristics for identifying the"importance"of a given layer towards the loss, based on assumptions such as \emph{error monotonicity}, i.e. that the end-to-end model compression error is proportional to the sum of layer-wise errors. In this paper, we revisit this area, and propose a new and general approach for dynamic compression that is provably optimal in a given input range. We begin from the motivating observation that, in general, \emph{error monotonicity does not hold for LLMs}: compressed models with lower sum of per-layer errors can perform \emph{worse} than models with higher error sums. To address this, we propose a new general evolutionary framework for dynamic LLM compression called EvoPress, which has provable convergence, and low sample and evaluation complexity. We show that these theoretical guarantees lead to highly competitive practical performance for dynamic compression of Llama, Mistral and Phi models. Via EvoPress, we set new state-of-the-art results across all compression approaches: structural pruning (block/layer dropping), unstructured sparsity, as well as quantization with dynamic bitwidths. Our code is available at https://github.com/IST-DASLab/EvoPress.

Zagovornici nekritičkog korištenja informatičke tehnologije u obrazovanju tvrde da je budućnost obrazovanja u kibernetskome prostoru i digitalizaciji. No, digitalna tehnologija nije alternativa za poželjnu interakciju. Zato se danas u suvremenoj pedagogiji i govori o socijalnome biću škole kao polazištu humane škole. Kritičkomu mišljenju treba posvetiti više pozornosti u odgoju i obrazovanju, a pretpostavka za razvoj kritičkoga mišljenja jest uvođenje toga kolegija na svim društveno-humanističkim fakultetima, a onda i kao izborni predmet u osnovnim školama. U tome kontekst predlažemo znatno šire od sadašnjih aspekte sadržaja medijske kulture u čitankama za hrvatski jezik i književnost, osobito u osnovnim školama. Ključne riječi: mediji; manipulacija; indoktrinacija; odgoj za kritičko mišljenje; medijska kultura.

This research delves into the dynamic shifts in human resource management strategies prompted by the COVID-19 pandemic. It investigates the mediating influence of information technology (IT) challenges on the connection between pandemic-induced international human resource management (IHRM) practices and the competitiveness of multinational corporations (MNCs) in Malaysia with focusing on sustainable development. Through the analysis of data collected from 172 respondents via self-administered questionnaires in Malaysian MNCs across various sectors including education, general services, ICT, property, construction, and healthcare, the study employs partial least squares structural equation modeling (PLS-SEM) to validate the proposed hypotheses. The findings highlight the substantial impact of compensation and staffing practices on technology transfer challenges within MNCs. Furthermore, the study reveals that the implementation of remote work, particularly during and post-lockdowns, is associated with elevated compensation and enhances overall company competitiveness. These outcomes offer theoretical and practical insights, furnishing human resource managers, especially in multinational corporations, with valuable guidance for maintaining competitiveness amidst the disruptions of a pandemic and promoting sustainability in HR practices. By highlighting the potential benefits of working from home in terms of both IHRM outcomes and competitiveness, the study contributes to ongoing discussions about the future of work and the role of technology-enabled practices in driving organizational success and sustainable development.

Effective real-time monitoring technique is crucial for detecting material degradation and maintaining the structural integrity of nuclear systems to ensure both safety and operational efficiency. Traditional physical sensor systems face limitations such as installation challenges, high costs, and difficulties in measuring critical parameters in hard-to-reach or harsh environments, often resulting in incomplete data coverage. Machine learning-driven virtual sensors offer a promising solution by enhancing physical sensor capabilities to monitor critical degradation indicators like pressure, velocity, and turbulence. However, conventional machine learning models struggle with real-time monitoring due to the high-dimensional nature of reactor data and the need for frequent retraining. This paper explores the use of Deep Operator Networks (DeepONet) within a digital twin (DT) framework to predict key thermal-hydraulic parameters in the hot leg of an AP-1000 Pressurized Water Reactor (PWR). In this study, DeepONet is trained with different operational conditions, which relaxes the requirement of continuous retraining, making it suitable for online and real-time prediction components for DT. Our results show that DeepONet achieves accurate predictions with low mean squared error and relative L2 error and can make predictions on unknown data 160,000 times faster than traditional finite element (FE) simulations. This speed and accuracy make DeepONet a powerful tool for tracking conditions that contribute to material degradation in real-time, enhancing reactor safety and longevity.

Real-time monitoring is a foundation of nuclear digital twin technology, crucial for detecting material degradation and maintaining nuclear system integrity. Traditional physical sensor systems face limitations, particularly in measuring critical parameters in hard-to-reach or harsh environments, often resulting in incomplete data coverage. Machine learning-driven virtual sensors offer a transformative solution by complementing physical sensors in monitoring critical degradation indicators. This paper introduces the use of Deep Operator Networks (DeepONet) to predict key thermal-hydraulic parameters in the hot leg of pressurized water reactor. DeepONet acts as a virtual sensor, mapping operational inputs to spatially distributed system behaviors without requiring frequent retraining. Our results show that DeepONet achieves low mean squared and Relative L2 error, making predictions 1400 times faster than traditional CFD simulations. These characteristics enable DeepONet to function as a real-time virtual sensor, synchronizing with the physical system to track degradation conditions and provide insights within the digital twin framework for nuclear systems.

The article presents the results obtained within the DaWetRest project, which focuses on wetlands preservation and restoration along the Danube River, ranging on three demonstrator sites: the Middle Danube Demo (MD DEMO located in Croatia), the Lower Danube Demo (LD DEMO, located in Bulgaria) and the Danube Delta Demo (DD DEMO-located in Romania). The project aims to provide insight into nature-based solutions, ecological characteristics, hydrology, water quality, vegetation, wildlife, soil, land use and other factors that represent a relevant contribution to the restoration efforts and it contributes to the wetland restoration using concreate hydro-technical measures for ecosystem improvement. This study presents the initial results from in situ applied geophysics, where Electric Resistivity Tomographies (ERT) were applied on the MD Demo (main pilot site in the Municipality of Draž, Croatia) which represents the flood area, located east of the Danube River embankment of the State border. This geophysical surveys represents an exploratory approach that aims to contribute for a better understanding and knowledge on groundwater resources and quality through the determination of potential aquifer areas from main geological features of the area of interest and also as a potential contribution to establish carbon sequestration areas.

Effective real-time monitoring technique is crucial for detecting material degradation and maintaining the structural integrity of nuclear systems to ensure both safety and operational efficiency. Traditional physical sensor systems face limitations such as installation challenges, high costs, and difficulties in measuring critical parameters in hard-to-reach or harsh environments, often resulting in incomplete data coverage. Machine learning-driven virtual sensors offer a promising solution by enhancing physical sensor capabilities to monitor critical degradation indicators like pressure, velocity, and turbulence. However, conventional machine learning models struggle with real-time monitoring due to the high-dimensional nature of reactor data and the need for frequent retraining. This paper explores the use of Deep Operator Networks (DeepONet) within a digital twin (DT) framework to predict key thermal-hydraulic parameters in the hot leg of an AP-1000 Pressurized Water Reactor (PWR). In this study, DeepONet is trained with different operational conditions, which relaxes the requirement of continuous retraining, making it suitable for online and real-time prediction components for DT. Our results show that DeepONet achieves accurate predictions with low mean squared error and relative L2 error and can make predictions on unknown data 160,000 times faster than traditional finite element (FE) simulations. This speed and accuracy make DeepONet a powerful tool for tracking conditions that contribute to material degradation in real-time, enhancing reactor safety and longevity.

The capture of a rare, critically endangered adult angular rough shark, Oxynotus centrina (Linnaeus, 1758), with abnormal coloration is reported in this paper. The shark exhibited a partial reduction in pigmentation, resulting in an overall pale appearance with white-greyish patches. Since the retinal pigmentation appeared normal, the shark was considered leucistic. This represents the first documented case of leucism in this species and the first colour disorder reported in the family Oxynotidae Gill, 1912. Despite the atypical appearance, the physical health of the shark seemed unaffected, supporting the notion that pigment disorders in deep-sea sharks do not inherently impair survival and growth. Full morphometric characteristics are presented and compared with those of a normal individual of the same sex caught in the same area, showing no differences.

The personality traits of top judo referees are crucial for fair decision-making in elite competitions, shaping the experience for athletes, coaches, and spectators. This study examines potential differences in personality traits among 63 referees from the World Judo Tour between 1 January 2018 and 31 December 2022. Factors analyzed include completing the IJF Academy course Level 1, elite athlete status, number of officiated events, performance ratings, and participation in the Olympic or Paralympic Games. Our research shows that older referees tend to exhibit greater extraversion, whereas less experienced officials show lower levels of this trait. Referees with limited experience generally demonstrate higher agreeableness than their more experienced counterparts. Female referees and those with top performance ratings display greater conscientiousness than male referees. Completing the IJF Academy course is associated with lower neuroticism, while lower performance ratings are linked to higher neuroticism. Openness tends to decrease with increased officiating experience, with less experienced referees showing higher levels of this trait. In conclusion, competitive experience, training completion, and officiating tenure are associated with specific personality traits among judo referees, highlighting the importance of continuous training for effective officiating. The analysis of personality traits revealed no statistically significant differences between male and female referees in the dimensions measured by the BFI (Big Five Inventory). This indicates that the levels of extraversion, agreeableness, conscientiousness, neuroticism, and openness were similar for both genders, with no significant variation in how these traits were expressed.

Automated vehicles communicating with each other or their surroundings are expected to exchange a large amount of data. With that, the trustworthiness of a shared data item concerning its integrity is raised, as well as the trustworthiness of a vehicle component not having been tampered with by an attacker. Traditional security mechanisms, such as misbehavior detection, can help identify some security violations but cannot assess the overall consequences of a range of vehicle attacks. For this purpose, previous work has already introduced the Trust Assessment Framework, which computes a target entity’s Actual Trustworthiness Level (ATL). This paper focuses on the concept of Required Trustworthiness Level (RTL), which represents the numerical thresholds an ATL needs to reach for an entity to be considered trustworthy. We present a risk-based method to calculate the belief component of an RTL based on the well-established and standardized Threat Analysis and Risk Assessment (TARA). We provide an in-vehicle use case to demonstrate our belief calculation method and discuss the impact of using risk ratings.

TrialsNet is a project dedicated to enhancing European urban ecosystems through a variety of innovative use cases in domains including Security and Safety, Infrastructure, and Transportation. These use cases are being implemented across different clusters in Italy, Spain, Greece, and Romania, involving real users. This paper provides an overview of the diverse use cases, and of the corresponding network solutions, which leverage advanced functionalities like dynamic slicing management, NFV, MEC, AI/ML, and more. The project aims to identify network limitations, optimize infrastructure, and define new requirements for next-generation mobile networks. Ultimately, TrialsNet seeks to improve urban livability by driving advancements across multiple domains.

Only the chairs can edit The integration of vehicular communications, 5G mobile networks, and edge computing represents a significant shift in intelligent transportation. Key components of Intelligent Transportation Systems, such as Vehicle-to-Vehicle and Vehicle-to-Infrastructure communications, are essential for this transformation. The introduction of 5G improves connectivity, while edge computing brings processing capabilities closer to data sources. This combination has the potential to dramatically enhance transportation efficiency and safety. We focus on developing a sustainable Vehicle-to-Everything (V2X) framework based on experimentation in the Smart Highway testbed, located in Antwerp, focusing on protecting Vulnerable Road Users (VRUs). This study explores the interaction between vehicular communication and edge computing within a 5G network, focusing on the varying distances between On Board Units (OBUs) and Roadside Units (RSUs). The framework applications involve the development of a VRU Safety Mobile Application (SMA) and a Collision Prediction Edge Application (CPEA). Additionally, the research addresses sustainability by analyzing energy consumption in the context of the Central Processing Unit (CPU) load at the RSU using detailed real-world experiments and simulations. The findings indicate that energy consumption remains stable at shorter distances but shows increased variability at longer ranges.

This study was undertaken to examine vestibulo‐ocular reflex (VOR) characteristics in myotonic dystrophy type 1 (DM1) and type 2 (DM2) using video head impulse testing (vHIT).