In this paper, a neuro-adaptive controller with weight norm constraints is proposed for uncertain Euler-Lagrange systems. The boundedness of the weights in the neuro-adaptive controller design is important to prevent excessively large control inputs and system instability. To ensure the boundedness of the weights, the weight norm constraints are imposed as inequality constraints in the weight adaptation. The adaptation law is derived based on the constrained optimization method. The stability of the proposed controller is analyzed in the sense of Lyapunov, ensuring the boundedness of the tracking error and weight estimation. For the comparative study, two benchmark controllers and the proposed controller were evaluated through a numerical simulation of a two-link manipulator system and compared in terms of tracking performance and parameter dependency. The comparative study verified that the proposed controller has better tracking performance and lower parameter dependency.

Further progress in industrial electrochemical generators' designs utilising SOFC considerably depends on problem of obtaining fuel cell materials with predefined properties [1,2]. Due to this fact, considerable efforts of designers are directed on creation of optimum composition «anode-electrolyte-cathode», and on microconstruction of SOFC materials (selection of phase and element composition, defect structure formation, etc.). In the second problem nanocomposite systems having unique properties are of greatest interest [3]. Utilising of such materials as solid electrolyte may allow significantly improve cell's parameters, in particular, increase ionic conductivity [4,5]. However mechanisms leading to this effect were not determined singly. In this article the influence of nanoparticles ensemble with work of exit that different from the rest of the matrix on the change of electron density in SOFC materials have been examined theoretically. The calculations of concentration distribution of free carriers around single particle and for microparticles ensemble were fulfilled. In conclusion possible directions of forming of nanostructures conformably to SOFC materials discussed.

A critical component for advancing driving automation is trajectory prediction, which enables vehicles to navigate complex driving scenarios and make informed decisions in unfamiliar environments. However, training models exclusively on trajectory supervision presents significant challenges for trajectory prediction in autonomous systems. We propose a Velocity-Angle Trajectory Network (VATNet), an integrated approach that enhances trajectory prediction tasks with an additional velocity prediction and angle prediction task in a multi-task learning framework. This assists the model in cross-validating and correcting errors in trajectory prediction, leveraging temporal relationships and directional context to achieve more reliable outcomes. To address the challenges posed by the periodic nature of angles, we use a classification approach. This method discretizes angles into bins and focuses on probability distributions rather than precise values, thereby reducing numerical instability associated with regression. To mitigate the issue of overfitting to any single task, we introduce an uncertainty-aware weighted loss that dynamically assigns higher priority to tasks identified as challenging or uncertain. The VATNet model has been evaluated on Argoverse 2 dataset and outperforms state-of-the-art models by a margin of 3.76%-33.83%, 5.68%-42.97%, 11.84%-90.78% and 0.37%-29.73% on the minADE6, minFDE6, MR6 and b-FDE6 respectively.

Grid-forming (GFM) inverters, like synchronous generators (SGs), form voltage at their terminals and are considered crucial for enabling high renewable energy integration. Most GFM controllers mimic the power-synchronization characteristics of the SG rotor's swing dynamics to synchronize with the system. However, GFM inverters differ significantly from SGs in both design and control. This paper systematically explores these differences. The analysis reveals that the coupling between steady-state and transient performance in traditional GFM controllers, both influenced by the active power-frequency droop gain, along with the absence of features like large stator inductance, damper windings, and power system stabilizers (PSS), may result in SGs having superior disturbance rejection in the small-signal domain. Building on these insights, a novel GFM control technique with improved disturbance rejection performance is proposed. A comparative stability analysis is performed to evaluate the robustness of the proposed technique against conventional GFM controls under various grid conditions. The effectiveness of the proposed control method is further validated through MATLAB/Simulink simulations and real-time hardware-in-the-loop testing in diverse power system scenarios.

Critical illness is a condition that requires special attention, as it often involves life-threatening situations with a higher mortality rate. Attention is further focused on diabetic patients, who have recently become common comorbid cases among critically ill patients. Patients with these conditions are often prone to hyperglycemia and hypoglycemia due to the nature of critical illness and diabetes itself. Intensive Insulin Therapy (IIT) is known to be beneficial in maintaining blood glucose levels at the optimal target in such patients. However, the application of IIT is not straightforward. To date, there is no standardized IIT protocol, and its use is highly dependent on the expertise of each clinician. Moreover, each protocol does not always guarantee the maintenance of blood glucose levels because it depends on the patient's response to insulin, while the protocol is designed to be general. With the rapid advancement of Artificial Intelligence (AI), particularly Machine Learning (ML), a personalized IIT can be developed to address this issue. With the power of data, AI algorithms can analyze and assist in creating personalized IITs according to the patient's condition, with the goal of maintaining 2 their blood glucose levels. In this review, we present a future directive on the potential role of AI in creating personalized IITs for critically ill diabetic patients, as well as the challenges to be overcome in the development process.

The increasing computational demands and inefficiencies associated with the scaling of advanced language models present significant challenges, particularly in terms of maintaining accuracy, inference speed, and resource utilization. Addressing these issues, the introduction of Dynamic Token Compression (DTC) and Adaptive Layer Pruning (ALP) offers a novel and transformative approach to optimizing model performance, significantly reducing token redundancy and selectively pruning model layers based on the complexity of input queries. These methods enable enhanced language models to generate more coherent, contextually accurate outputs while minimizing computational load, making them more suitable for real-time applications. Through comprehensive experimentation, DTC and ALP were shown to improve model efficiency across multiple key metrics, including perplexity, BLEU score, and hallucination rate, while also reducing memory usage and energy consumption. The ability to dynamically adjust token processing and layer utilization without compromising on performance demonstrates the practical value of these techniques for large-scale model deployments in various domains. By optimizing both computational and linguistic aspects of the models, this research provides a robust framework for future advancements in language model architecture.

6G networks aim to facilitate the global sustainability drive. However, sustainability of 6G networks itself will pose major research questions due to the increasing number of devices, systems, and services that can exacerbate the consumption of key resources. In this work, we focus our investigation to 6G security, since sustainability is mostly overlooked when it comes to security. Emerging security technologies which are highly dependent on developing concepts in the realm of security such as AI and blockchain require huge amounts of resources. This article aims to stir further research on sustainability of 6G security through highlighting the main research challenges and future research directions in this domain.

The escalating complexity of cyber threats requires increasingly sophisticated approaches to automated detection, particularly within the domain of ransomware. Existing methods, such as signature-based and heuristic detection, often fail to recognize novel ransomware variants due to their reliance on predefined, static indicators or anomaly-based heuristics, leading to high false positive and false negative rates. DeepCodeLock addresses these limitations through a deep learning framework that leverages convolutional and recurrent neural network layers to analyze ransomware behavioral patterns comprehensively, allowing for the extraction of high-level features indicative of malicious intent. By processing system activities like file access, network connections, and process hierarchy structures, the model autonomously generates a dynamic, composite behavioral signature for each monitored activity, enhancing detection precision. Experimental results demonstrated that DeepCodeLock achieved significant improvements in detection accuracy, reduced false positive and negative rates, and maintained low latency and resource utilization across varied network sizes, establishing its practicality for real-time, large-scale cybersecurity applications. Overall, DeepCodeLock's methodological innovations emphasize its utility as an adaptive and scalable ransomware detection system, offering new possibilities in protecting critical infrastructures from emergent cyber threats.

This paper presents our participation in the Capsule Vision 2024 Challenge, where we leverage FasterViT, a high-speed variant of the Vision Transformer (ViT), to address the problem of efficient image classification under resource constraints. By adapting FasterViT with targeted data augmentation, fine-tuning, and optimization strategies, we achieved improved accuracy and computational efficiency over standard baselines. Our approach demonstrates the potential of efficient transformers in real-world visual recognition applications.

Ultrasound molecular imaging (UMI) uses targeted microbubbles (MBs) to detect disease-associated biomarkers. For UMI, distinguishing the acoustic signals produced by bound MBs from those by free MBs and tissue is critical. Currently, the main approach, known as differential Targeted Enhancement (DTE), is time-intensive and requires MB destruction. Here we introduce a novel, rapid, and non-destructive UMI technique utilizing high-order singular value decomposition (HOSVD). HOSVD decomposes the signals of an acoustic contrast sequence, separating them owing to their nonlinear content and temporal coherence. The nonlinear separation enables distinction between tissue and MBs, while the temporal separation enables distinction between free and bound MBs. To determine the molecular signal, we defined a bound MB indicator 𝝌, calculated from the HOSVD output. The value of 𝝌, both in in silico and in vitro experiments, was consistently higher for bound MBs compared to free MBs and tissue, up to concentrations of 20x10 3 free MBs/ml. In addition, the molecular signal determined from 𝝌 correlated well with a DTE ground truth acquisition. The method was compared to other nondestructive techniques such as low-pass filtering and normalized singular spectrum area, demonstrating an average molecular signal enhancement of 12.1 dB. Furthermore, when used as a binary classifier, our method achieved a detection of up to 1.81× more true positives while reducing false positives up to 1.78×. These findings suggest that HOSVD could pave the way to rapid, nondestructive UMI.

Within the color space of 3-coloring, we build deterministic algorithms for plane graphs with no more than 4 to find base configurations that appear in the n-DJC group, and thus find new triplets bringing 3-inseparable vertices. This paper provides the basic theory of the n-DJC group, including the removable conflict group between DJC and the rotation of DJC in the group. We can identify 3-uncolorable vertices according to unreasonable removable conflicts or frozen rotations. And we point out that such n-DJC groups can appear in the color space of planar graph with degree no more than 4.

The rapid proliferation of digital content and the increasing complexity of language data have posed significant challenges for effective understanding and generation of text within computational frameworks. Novel approaches to optimizing token processing and reducing hallucination occurrences hold substantial promise for enhancing the coherence and contextual accuracy of machine-generated outputs. A comprehensive exploration of dynamic token prioritization offers a transformative solution to longstanding inefficiencies, enabling models to intelligently allocate computational resources in accordance with the contextual relevance of input data. By implementing an adaptive mechanism that continuously re-evaluates token significance during the inference process, considerable improvements in processing speed and output quality have been achieved. This research illustrates how sophisticated optimization techniques can refine the interaction between machine learning algorithms and linguistic data, fostering a more intuitive and reliable user experience. As the landscape of digital communication evolves, the implications of these advancements extend far beyond traditional applications, paving the way for innovative integrations across various industries, including healthcare, education, and content creation.

Local electricity markets are increasingly seen as a key factor in democratising the energy system. While there has been a significant amount of research on the technical and economic aspects of these markets, it is still not clear how well they align with the social acceptance of potential participants. The concept of fairness in these markets has been gaining attention, but there has not been a clear discussion or definition of what constitutes fairness in a local electricity market context. Furthermore, some fairness indicators have become popular in the literature on local electricity markets, but have yet to be fully analysed on how well they capture fairness in these markets. This study aims to address these issues by formally defining distributive energy justice in the context of local electricity markets and evaluating how well popular fairness indicators perform based on this definition. The results of this extensive evaluation lead to proposed adjustments to the indicators to make them more suitable for local electricity markets, and a discussion on future research directions for fairness in local electricity markets.

Objective: Using a 3-degree-of-freedom (DOF) prosthetic foot with lateral and toe joint compliance, this study aims to evaluate the combined effects these features in prosthetic gait through gait trials on 10 non-amputated individual. Results: The addition of coronal and toe compliance had significant kinematic and muscular effects. Notably, this compliance combination reduced peak pelvis obliquity by 27%, preserved the swing stance/ratio, and decreased gluteus medius’s activation by 34% on the non prosthetic side, compared to the laterally rigid version of the prosthesis without toe compliance. Conclusions: The results underscore the importance of integrating coronal and toe compliance in prosthetic feet designs as they shows potential in improving gait metrics related to mediolateral movements and balance, while also decreasing muscle activation. Still, these findings remain to be validated in people with transtibial amputations.

The increasing complexity of modern cyber threats requires the development of more advanced and dynamic detection systems, capable of adapting to new and evolving attack vectors. Dynamic Behavior Trace Profiling (DBTP) introduces a novel method for ransomware detection, utilizing real-time behavior analysis to identify malicious activities without relying on static signatures. Through continuous monitoring of systemlevel events, such as file I/O operations, network anomalies, and process manipulation, DBTP constructs detailed behavioral profiles that distinguish between benign and malicious activities. The system's modular architecture enables it to efficiently handle large volumes of data while maintaining low detection latency, making it suitable for real-time deployment in diverse operational environments. Experimental results highlight DBTP's high detection accuracy, particularly in identifying file-based ransomware activities, while also demonstrating its adaptability to previously unseen ransomware variants. False positive and false negative rates remain low across different test scenarios, indicating the system's reliability. Moreover, DBTP's capacity to function under high system load conditions with minimal resource overhead positions it as an effective solution for modern cybersecurity infrastructures. Overall, DBTP provides a scalable, automated, and adaptive tool for detecting and mitigating ransomware threats, significantly enhancing the resilience of systems against emerging and sophisticated ransomware attacks.

Alzheimer's Disease (AD) is a prevalent neurodegenerative condition where early detection is vital. Handwriting, often affected early in AD, offers a non-invasive and costeffective way to capture subtle motor changes. State-of-the-art research on handwriting, mostly online, based AD detection has predominantly relied on manually extracted features, fed as input to shallow machine learning models. Some recent works have proposed deep learning (DL)-based models, either 1D-CNN or 2D-CNN architectures, with performance comparing favorably to handcrafted schemes. These approaches, however, overlook the intrinsic relationship between the 2D spatial patterns of handwriting strokes and their 1D dynamic characteristics, thus limiting their capacity to capture the multimodal nature of handwriting data. Moreover, the application of Transformer models remains basically unexplored. To address these limitations, we propose a novel approach for AD detection, consisting of a learnable multimodal hybrid attention model that integrates simultaneously 2D handwriting images with 1D dynamic handwriting signals. Our model leverages a gated mechanism to combine similarity and difference attention, blending the two modalities and learning robust features by incorporating information at different scales. Our model achieved state-of-theart performance on the DARWIN dataset, with an F1-score of 90.32% and accuracy of 90.91% in Task 8 ('L' writing), surpassing the previous best by 4.61% and 6.06% respectively.

The "No Free Lunch" (NFL) theorem in machine learning asserts that there is no universal algorithm or model that excels across all possible problem instances. In the context of machine learning applications for stock market prediction, this theorem implies that no single predictive model can consistently outperform others under all market conditions. The NFL theorem, introduced by David Wolpert in 1996, emphasizes the importance of considering the specific characteristics of the problem domain. In this paper, we extend the NFL Theorem as "No Free Money(NFM) Theorem for Machine learning applications in stock market predictions. The data gathered from Machine Learning competition for stock market prediction are utilized for providing experimental evidence for NFM theorem. For stock market prediction, the NFM theorem underscores the need for tailored approaches, acknowledging that the effectiveness of machine learning models is contingent on factors such as market dynamics, economic conditions, and the quality of historical data. It cautions against assuming a one-size-fits-all solution and highlights the challenge of developing models that generalize well to diverse and evolving market scenarios. The theorem prompts practitioners to approach stock market prediction with a realistic understanding of the limitations inherent in algorithmic approaches, encouraging careful consideration of the data, features, and context relevant to each specific application.

Cellular heterogeneity is one of the main hallmarks of cancer, referring to the coexistence of different phenotypes with very distinct biological behaviors in single isolates. Automatically detecting single-cell heterogeneity is therefore critical, and can provide important information on cancer initiation, tumour aggressiveness or drug-related resistance. Here we introduce He2Cl, a novel machinelearning algorithm able to detect single cell heterogeneity from the morpho-dynamic analysis of 2D cell cultures. Built on label-free quantitative time-lapse imaging, He2Cl performs a 2-step clustering algorithm for unsupervised classification of sub-phenotypes within a cell population. He2Cl outperforms state-of-the-art clustering models and allows the simultaneous classification of thousands of cell trajectories from their birth to their division, paving the way towards AI-enhanced microscopy for live-cell analysis.