On the threshold of a new technological era, Sixth Generation (6G) networks promise to revolutionize global connectivity, bringing mobile communications to data speeds in the terabits per second range and ultra-low latency. These networks will enhance the user experience enable a wide range of advanced applications and emerging services. Artificial Intelligence (AI)-powered network functions and services, also known as Network Intelligence Functions (NIF) and Network Intelligence Service (NIS), are essential to achieve this vision. In this study, we present the design and development of an end-to-end framework for orchestrating AI-based functions. Utilizing Kubernetes (K8s) and Prefect, we showcase its implementation through an AI-driven Traffic Classification (TC) use case. Our results confirm the feasibility of the proposed framework, offering valuable insights in the lifecycle management design, such as data collection, decision-making, and critical performance metrics, including deployment time and model performance in terms of accuracy and inference times among three different Machine Learning (ML)-based TC models.

The increasing demand for high-quality and efficient Channel Estimation (CE) in 5G New Radio (5G-NR) systems has prompted the exploration of advanced Deep Learning (DL) techniques. While traditional methods, such as Linear Interpolation (LI) and Least Squares (LS), provide reasonable accuracy and are practical for real-time physical layer processing, recent DL-based CE approaches have primarily focused on accuracy, often without evidence of real-time capabilities. In this paper, we present a comprehensive evaluation of DL-based Super-resolution (SR) methods for CE, comparing models like Super Resolution Convolutional Neural Network (SRCNN), ChannelNet, and Enhanced Deep Super-Resolution (EDSR) in both 1D and 2D convolutional architectures. We optimize these models using NVIDIA TensorRT to reduce computational complexity and latency. Our results show that the optimized 1D-EDSR model achieves the best performance with a Mean Squared Error (MSE) of 0.0126, outperforming all other models in terms of accuracy. However, the optimized 1D-EDSR model fails to meet real-time constraints due to additional computational overhead (0.6798 ms/sample). In contrast, the 1D-SRCNN model offers a balanced trade-off between MSE (0.01738) and inference time (0.0866ms/sample), achieving 40% higher accuracy than LS (0.0288) while maintaining the best energy efficiency (1.48 mJ/sample).

Image-based high-throughput plant phenotyping utilises various imaging techniques to automatically and non-invasively understand the growth of different plant species. These innovative imaging infrastructures are implemented to monitor plant development over time in indoor or outdoor environments. However, understanding the relationship between genotype and phenotype interactions under different environments remains challenging. This research study demonstrates superior extraction of leaf morphological features of different Arabidopsis thaliana ecotypes by analysing leaf geometry using a sequence of RGB images. Upon successful extraction of anatomical features, leaf length and area are converted into physical coordinates. Furthermore, considering these leaf features as 1D signals, the Fourier Spectrum is analysed, and most descriptive features are selected using PCA. Finally, leaf shape classification is established by training and testing five distinct ML models. A thorough evaluation of selected models demonstrates superiority in classifying two common leaf shapes of Arabidopsis plants.

This study describes a method for measuring flexion and extension motions in the hand, which is crucial for healing and restoring a complete range of motion following accidents. Precise control is essential because of the human hand's intricate anatomy, which consists of various bones and joints. The method replicates the trajectories of the five fingers, emphasizing their position, speed, acceleration, and torque at four crucial structural points: four fingers that stretch to 90 degrees and then return to 0 degrees, with the thumb creating a 20-degree angle, a crucial initial position for successful recovery. The ability to simulate these trajectories is essential for correctly tracking patients' progress and customizing rehabilitation treatments, both of which improve patient care and rehabilitation outcomes.

Increased integration of photovoltaic (PV) production creates new challenges and opportunities for the Distribution System Operators (DSO). It is crucial to establish comprehensive standards for the maximum integration of PV systems without disrupting normal operations. This paper explores the hosting capacity of concentrated and dispersed PV systems connected to LV network. The aim is to determine the maximum amount of PV generation for the low-voltage network considering voltage, line loads and transformer thermal constraints. Simulations are performed on a model of real power systems in DigSilent Power Factory software. This paper encompasses a wide range of activities vital for the successful integration of PV systems into low voltage distribution network that has been converted into micro-grid. Goal was to ensure stability, security, and efficiency of the grid while exploring best possible ways of connecting new renewable energy sources.

In the literature, auditory attention is explored through neural speech tracking, primarily entailing modeling and analyzing electroencephalography (EEG) responses to natural speech via linear filtering. Our study takes a novel approach, introducing an enhanced coherence estimation technique to assess the strength of neural speech tracking. This enables effective discrimination between attended and ignored speech. To mitigate the impact of colored noise in EEG, we address two biases–overall coherence-level bias and spectral peak-shifting bias. In a listening study involving 32 participants with hearing impairment, tasked with attending to competing talkers in background noise, our coherence-based method effectively discerns EEG representations of attended and ignored speech. We comprehensively analyze frequency bands, individual frequencies, and EEG channels. Frequency bands of importance are shown to be delta, theta and alpha, and the important EEG channels are the central. Lastly, we showcase coherence differences across different noise reduction settings implemented in hearing aids (HAs), underscoring our method's potential to objectively assess auditory attention and enhance HA efficacy.

Background: Graft vs. host disease (GVHD) remains a major complication of allogeneic hematopoietic stem cell transplantation (alloHSCT). To create space for donor stem cells and prevent their rejection, alloHSCT protocols rely on conditioning regimens involving chemotherapy and radiation. Conditioning causes tissue damage, which increases the tissue injury signal or “alarmin” interleukin (IL)-33 in fibroblastic reticular cells (FRC) of the secondary lymphoid organs (SLO). Mechanisms releasing IL-33 from its sequestration in the nucleus remain elusive, but free IL-33 directly stimulates donor CD4 T cells to prime IL-12-independent Type 1 T helper cell (Th1) differentiation and expansion. Targeting IL-33 early after alloHSCT limits GVHD in pre-clinical models. The gastrointestinal tract (GIT) also upregulates IL-33 in response to TBI and GVHD, but a direct role for local IL-33 in sustaining pathogenic donor responses is unclear. Our goal was to manipulate the IL-33 pathway in the SLO or GIT to better understand how stromal communications with donor T cells initiate and shape GVHD and graft vs. lymphoma (GVL) responses. Methods: We compared donor T cells (plus or minus inducible deletion of the IL-33 receptor, ST2) for their ability to mediate GVHD vs. GVL (A20 lymphoma) in BALB/c recipients receiving total body irradiation (TBI) and CD45.1+ B6 T cell depleted bone marrow (TCD BM). To define the role for IL-33-derived from the SLO vs. the GIT, we assessed survival of B6 recipients deficient in IL-33 in FRCs (CCL19-CrexIl33fl/fl) vs. those deficient in IL-33 in the epithelium of the GI tract (Vil-CrexIl33fl/fl) receiving TBI and BALB/c T cells. To investigate if donor T cells mediate IL-33 release, we completed an ex vivo model using B6 St2+/+, St2-/-, and GzmB-/- CD3 T cells co-cultured for 5 days with BALB/c TCD splenocytes and LN-derived FRCs that had been irradiated at 3500 cGy alone or with the IL-33 antagonist, sST2. Similar in vivo studies were conducted where the above donor B6 St2+/+, St2-/-, and GzmB-/- CD3 T cells were transplanted into BALB/c recipients and assessed for GzmB and donor T cell expansion on day 5 post-alloHSCT. Results: Ablating ST2 at days 10-14 post-transplant (after initial GVHD development) improved clinical scores and limited mortality. Further, sustained IL-33 signaling was not required for GVL activity. Mechanistically, late ST2 deletion was associated with increased Foxp3 expression and reciprocal Tbet decrease in donor CD4+ T cells from both SLO and GVHD target tissues. Sustained IL-33 signaling also maintained donor T cell TCF1 expression in SLO. Surprisingly, isolated deletion of FRC-derived IL-33 increased GVHD mortality in the CCL19-CrexIl33fl/flrecipients. Mechanistic studies showing FRC-derived IL-33 stimulated CD4+ PD-1 expression and blunted the total number of CD4 and CD8 T effectors in the GIT at day 21 post-alloHSCT. Whereas, deletion of IL-33 in the gut epithelium in the Vil-CrexIl33fl/fl recipients was protective and prolonged survival. RNAseq analysis suggested that IL-33 stimulates T cell granzyme B (GzmB) expression. GzmB deficient (Gzmb-/-) donor T cells displaying reduced activation and expansion in vitro and in vivo, in a phenotype similar to ST2 deficient CD4 T cells. Consistent with the importance of GzmB in mediating IL-33 signals, antagonizing IL-33 had no impact on GzmB-/- T cell responses similar to ST2 deficient CD4 T cells when compared to Gzmb+/+, which failed to expand when IL-33 was sequestered. Conclusions: Our data reveals that GzmB-mediated crosstalk between donor T cells and IL-33+ stroma orchestrates donor T cell identities and tunes local alloimmune responses after alloHSCT. Delayed deletion of ST2 signaling on donor T cells promotes survival through an upregulation of regulatory mechanisms in GVHD target tissues. Similarly, targeted deletion of IL-33 in the GIT provides protection from donor driven pathology. Whereas, targeted deletion of IL-33 from SLO FRC promotes GVHD mortality by down regulating intrinsic T cell exhaustion mechanisms in the SLO, which impacts later CD4+ T cell alloimmune responses to available IL-33 in target tissues, driving GVHD pathology. These data suggest distinct temporal and tissue specific roles for IL-33-driven programing of donor CD4+ T cells. In total, these data indicate that continual feedback between donor T cells and recipient stroma is central to the development and maintenance of GVHD.

5G Standalone (SA) networks introduce a range of new applications, including enhanced Mobile Broadband (eMBB), Ultra-Reliable Low-Latency Communication (URLLC), and massive Machine-Type Communications (mMTC). Each of these applications has distinct network requirements, which current commercial network architectures, such as 4G and 5G Non-Standalone (NSA), struggle to meet simultaneously due to their one-size-fits-all design. The 5G SA architecture addresses this challenge through Network Slicing, creating multiple isolated virtual networks on top a single physical infrastructure. Isolation between slices is crucial for performance, security, and reliability. Each slice owns virtual resources, based on the physical resources (e.g., CPU, memory, antennas, and network interfaces) shared by the overall infrastructure. To deploy Network Slicing, it is essential to understand the concept of isolation. The Third Generation Partnership Project (3GPP) is standardizing security for Network Slicing, focusing on authentication, authorization, and slice management. However, the standards do not clearly define the meaning of isolation and its implementation in the infrastructure layer.In this paper, we define and showcase a real-life Proof of Concept (PoC), which guarantees isolation between slices in 5G SA networks, for each network domain i.e., Radio Access Network (RAN), Transport Network (TN), and 5G Core (5GC) network. Furthermore, we describe the 5G SA architecture of the PoC, explaining the isolation concepts within the Network Slicing framework, how to implement isolation in each network domain, and how to evaluate it.

Introduction Techniques for assessing the blood plasma proteome with high precision and at great depth are rapidly developing and have demonstrated utility in carrying diagnostic and prognostic information for patients with cancer, including hematological malignancies. However, it is not known whether the plasma proteome can be useful in distinguishing the more closely related cancer entities, such as different B-cell lymphomas (BCLs). Performing affinity-based plasma proteomics analyses in a population-based cohort of BCLs, we aimed at discovering plasma proteome differences between BCL subtypes and identifying potential biomarkers that can aid differential diagnosis. Material and Methods We analyzed 592 BCLs (221 diffuse large BCL (DLBCL), 94 follicular lymphoma (FL), 123 Hodgkin lymphoma (HL), 91 mantle cell lymphoma (MCL), and 63 primary CNS lymphoma (PCNSL)) from the U-CAN biobank (www.u-can.uu.se). Plasma samples collected at diagnosis were analyzed using the Olink Explore 1536 platform, which provided relative quantification of 1463 unique proteins. The plasma proteomes between a given group and all the remaining groups were compared with a two-sided t test and further adjusted for age and sex in multivariable linear limma models. To identify panels of plasma proteins that can differentiate between the different subtypes of BCLs, we trained two types of machine learning (ML) models based on the random forest (RF) algorithm and logistic regression with regularization (LRR). The entire dataset was proportionally partitioned into a training (70%) and testing (30%) dataset. Both model types were trained in one thousand iterations, with cross-validation, on a non-filtered dataset and implementing different filtering approaches based on varying cut-offs of mean log2-difference (log2-diff) of differentially altered proteins (DAPs) and 0.1% false discovery rate (FDR). Finally, the best-performing model from the iterations of the two ML methods on the training data was selected and tested on the testing dataset for performance. Both balanced accuracy and area under the curve (AUC) were considered as main outcomes of performance. Results Comparing the plasma proteomes between BCL subtypes showed many DAPs in each subtype compared to the rest of the cohort at 5% FDR. PCNSL patients had the largest number of DAPs, followed by HL, MCL, DLBCL, and FL. However, most of these alterations were of smaller log2-diff between the subgroups. Less than ten proteins per group had a log2-diff > 1 in a subgroup compared to other subtypes, apart from MCL patients, who had 64 DAPs with log2-diff > 1. The findings remained consistent in the multivariable analyses, where the log2-diff between subgroups was adjusted for age and sex. Yet, each subgroup had more DAPs that were uniquely altered in that subgroup and in no other group, regardless of the log2-FC, with most DAPs observed again in the MCL, followed by DLBCL, HL, FL, and PCNSL. This was reflected in the ML models, where combining smaller differences in protein levels into multivariate models showed reliable performance in differentiating the BCLs. Filtering improved the model's accuracy, and the derived best-performing LRR model showed moderate to high accuracy in differentiating the BCLs on testing data. The LRR model had the highest accuracy in classifying MCL, with AUC of 91%, followed by HL (90%), PCNSL (89%), DLBCL (85%), and FL (80%), the latter being repeatedly misclassified in the ML iterations. Although the model's sensitivity was variable, being highest for HL and lowest for FL, the specificity was very high (>93%) for excluding FL (94%), HL (96%), MCL (98%), and particularly PCNSL (99%), with the negative predictive value of the model for CNS involvement being 98%. Conclusions Plasma proteomics can differentiate between distinct types of BCLs with a moderate to high accuracy, between 80% and 91%. The models showed the highest accuracy in classifying MCL, likely due to the highest number of unique DAPs and proteins with large log2-diff observed in this subtype On average, the models showed better specificity, which is highly relevant for DLBCL, where a blood biomarker can serve as a quick diagnostic tool for initial exclusion of CNS involvement in a patient, with very high predictive value. This suggests that plasma proteomics could assist in the differential diagnosis of B-cell lymphomas and potentially for CNS-involvement.

5G Standalone (SA) networks introduce a range of new applications, including enhanced Mobile Broadband (eMBB), Ultra-Reliable Low-Latency Communication (URLLC), and massive Machine-Type Communications (mMTC). Each of these applications has distinct network requirements, which current commercial network architectures, such as 4G and 5G Non-Standalone (NSA), struggle to meet simultaneously due to their one-size-fits-all design. The 5G SA architecture addresses this challenge through Network Slicing, creating multiple isolated virtual networks on top of a single physical infrastructure. Isolation between slices is crucial for performance, security, and reliability. Each slice owns virtual resources, based on the physical resources (e.g., CPU, memory, antennas, and network interfaces) shared by the overall infrastructure.In this demo, we define and showcase a real-life Proof of Concept (PoC), which enables Network Slicing guaranteeing isolation between slices in 5G SA networks, for each network domain i.e., Radio Access Network (RAN), Transport Network (TN), and 5G Core (5GC) network.

Lavandula species are one of the most popular aromatic plants in the world and have a high content of high-quality essential oil (EO). Although there are many species in this genus, only lavandin (Lavandula intermedia Emeric ex Loisel.) and lavender (L. angustifolia Mill.)  are highly valued worldwide. The quality and yield of lavandin and lavender depend on genetic factors, environmental conditions and cultivation methods. Therefore, the aim of this paper is to research the effects of the application of biostimulant on the inflorescence yield and the quality of lavandin and lavender. The treatments used in this research consisted of a combination of different species (lavandin and lavender) and biostimulant (applied and unapplied). The research results show that all the research traits significantly depended on the used species and the applied biostimulant. The inflorescence yield, the content of total flavonoids, and the content of EO were higher in the lavandin species (477.3 g plant-1, 17.21 mg CAE g-1, 8.57 mL 100 g-1, respectively) than in the lavender species (180.5 g plant-1, 13.41 mg CAE g-1, and 3.69 mL 100 g-1, respectively). EOs of lavandin and lavender were rich in linalool and linalyl acetate. The use of biostimulators had a positive effect on the inflorescence yield and the content of essential oil. Furthermore, the applied biostimulant increased the linalool content in the essential oil of both researched species, i.e. it positively affected its quality.

Spherically symmetric Einstein-{\ae}ther (E{\AE}) theory with a Maxwell-like kinetic term is revisited. We consider a general choice of the metric and the \ae{}ther field, finding that:~(i) there is a gauge freedom allowing one always to use a diagonal metric; and~(ii) the nature of the Maxwell equation forces the \ae{}ther field to be time-like in the coordinate basis. We derive the vacuum solution and confirm that the innermost stable circular orbit (ISCO) and photon ring are enlarged relative to general relativity (GR). Buchdahl's theorem in E\AE{} theory is derived. For a uniform physical density, we find that the upper bound on compactness is always lower than in GR. Additionally, we observe that the Newtonian and E\AE{} radial acceleration relations run parallel in the low pressure limit. Our analysis of E\AE{} theory may offer novel insights into its interesting phenomenological generalization: \AE{}ther--scalar--tensor theory ({\AE}ST).

Atrial cardiomyopathy is closely associated with atrial fibrillation (AF), and some patients exhibit no dysfunction at rest but demonstrate evident changes in left atrial (LA) function and LA volume during exercise. This study aimed to identify distinguishing signs during exercise stress echocardiography (ESE) among patients in sinus rhythm (SR), with and without history of paroxysmal/persistent AF (PAF). A prospective cohort of 1055 patients in SR was enrolled across 12 centers. The main study cohort was divided into two groups: the modeling group (n = 513) and the verification group (n = 542). All patients underwent ESE, which included B-lines, LA volume index (LAVi), and LA strain of the reservoir phase (LASr). Age, resting and stress LAVi and LASr, and B-lines were identified as a combination of detectors for PAF in both groups. In the entire cohort, aside from resting and stress LAVi and LASr, additional parameters differentiating PAF and non-PAF patients were the presence of systemic hypertension, exercise E/e’ > 7, worse right ventricle (RV) contraction during exercise (∆ tricuspid annular plane systolic excursion < 5 mm), a lower left ventricular contractile reserve (< 1.6), and a reduced chronotropic reserve (heart rate reserve < 1.64). The composite score, summing all 9 items, yielded a score of > 4 as the best sensitivity (79%) and specificity (65%). ESE can complement rest echocardiography in the identification of previous PAF in patients with SR through the evaluation of LA functional reservoir and volume reserve, LV chronotropic, diastolic, and systolic reserve, and RV contractile reserve. A scoring system predicting the probability of PAF. The score was computed using the cutoff values as in the illustration. The score >4 demonstrated a sensitivity of 79% and a specificity of 65% of PAF.

Atherosclerotic cardiovascular disease (ASCVD) and consequent acute coronary syndromes (ACS) are substantial contributors to morbidity and mortality across Europe. Fortunately, as much as two thirds of this disease’s burden is modifiable, in particular by lipid-lowering therapy (LLT). Current guidelines are based on the sound premise that, with respect to low-density lipoprotein cholesterol (LDL-C), “lower is better for longer”, and recent data have strongly emphasised the need for also “the earlier the better”. In addition to statins, which have been available for several decades, ezetimibe, bempedoic acid (also as fixed dose combinations), and modulators of proprotein convertase subtilisin/kexin type 9 (PCSK9 inhibitors and inclisiran) are additionally very effective approaches to LLT, especially for those at very high and extremely high cardiovascular risk. In real life, however, clinical practice goals are still not met in a substantial proportion of patients (even in 70%). However, with the options we have available, we should render lipid disorders a rare disease. In April 2021, the International Lipid Expert Panel (ILEP) published its first position paper on the optimal use of LLT in post-ACS patients, which complemented the existing guidelines on the management of lipids in patients following ACS, which defined a group of “extremely high-risk” individuals and outlined scenarios where upfront combination therapy should be considered to improve access and adherence to LLT and, consequently, the therapy’s effectiveness. These updated recommendations build on the previous work, considering developments in the evidential underpinning of combination LLT, ongoing education on the role of lipid disorder therapy, and changes in the availability of lipid-lowering drugs. Our aim is to provide a guide to address this unmet clinical need, to provide clear practical advice, whilst acknowledging the need for patient-centred care, and accounting for often large differences in the availability of LLTs between countries.

Despite the popularity of large language model (LLM) quantization for inference acceleration, significant uncertainty remains regarding the accuracy-performance trade-offs associated with various quantization formats. We present a comprehensive empirical study of quantized accuracy, evaluating popular quantization formats (FP8, INT8, INT4) across academic benchmarks and real-world tasks, on the entire Llama-3.1 model family. Additionally, our study examines the difference in text generated by quantized models versus their uncompressed counterparts. Beyond benchmarks, we also present a couple of quantization improvements which allowed us to obtain state-of-the-art accuracy recovery results. Our investigation, encompassing over 500,000 individual evaluations, yields several key findings: (1) FP8 weight and activation quantization (W8A8-FP) is lossless across all model scales, (2) INT8 weight and activation quantization (W8A8-INT), when properly tuned, incurs surprisingly low 1-3% accuracy degradation, and (3) INT4 weight-only quantization (W4A16-INT) is competitive with 8-bit integer weight and activation quantization. To address the question of the"best"format for a given deployment environment, we conduct inference performance analysis using the popular open-source vLLM framework on various GPU architectures. We find that W4A16 offers the best cost-efficiency for synchronous deployments, and for asynchronous deployment on mid-tier GPUs. At the same time, W8A8 formats excel in asynchronous"continuous batching"deployment of mid- and large-size models on high-end GPUs. Our results provide a set of practical guidelines for deploying quantized LLMs across scales and performance requirements.