Leveraging recent advancements in machine learning-based flavor tagging, we develop an optimal analysis for measuring the hadronic cross-section ratios R_bRb, R_cRc, and R_sRs at the FCC-ee during its WWWW, ZhZh, and t\bar{t}tt‾ runs. Our results indicate up to a two-order-of-magnitude improvement in precision, providing an unprecedented test of the SM. Using these observables, along with R_\ellRℓ and R_tRt, we project sensitivity to flavor non-universal four-fermion (4F) interactions within the SMEFT, contributing both at the tree-level and through the renormalization group (RG). We highlight a subtle complementarity with RG-induced effects at the FCC-ee’s ZZ-pole. Our analysis demonstrates significant improvements over the current LEP-II and LHC bounds in probing flavor-conserving 4F operators involving heavy quark flavors and all lepton flavors. As an application, we explore simplified models addressing current BB-meson anomalies, demonstrating that FCC-ee can effectively probe the relevant parameter space. Finally, we design optimized search strategies for quark flavor-violating 4F interactions.

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.

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).

Quantization is a powerful tool for accelerating large language model (LLM) inference, but the accuracy-performance trade-offs across different formats remain unclear. In this paper, we conduct the most comprehensive empirical study to date, evaluating FP8, INT8, and INT4 quantization across academic benchmarks and real-world tasks on the entire Llama-3.1 model family. Through over 500,000 evaluations, our investigation yields several key findings: (1) FP8 (W8A8-FP) is effectively lossless across all model scales, (2) well-tuned INT8 (W8A8-INT) achieves surprisingly low (1-3\%) accuracy degradation, and (3) INT4 weight-only (W4A16-INT) is more competitive than expected, rivaling 8-bit quantization. Further, we investigate the optimal quantization format for different deployments by analyzing inference performance through the popular vLLM framework. Our analysis provides clear deployment recommendations: W4A16 is the most cost-efficient for synchronous setups, while W8A8 dominates in asynchronous continuous batching. For mixed workloads, the optimal choice depends on the specific use case. Our findings offer practical, data-driven guidelines for deploying quantized LLMs at scale -- ensuring the best balance between speed, efficiency, and accuracy.

This paper introduces a control system for Doubly Fed Induction Generator (DFIG) based on a Disturbance Observer (DOB) for island mode operation. The proposed control system is validated through experiments, confirming its effectiveness in maintaining stable operation during island mode. The system responded efficiently to variations in wind speed and load conditions, demonstrating the efficacy of the implemented control scheme. The proposed control system unifies the design approach for both the inner and outer loops of the cascaded control system structure, simplifying implementation and parameter tuning.

(1) Background: This case study analyzed the successful performances of female gymnasts in the finals of the 39th and 40th World Cup in Maribor (SLO). The aim was to identify variations in their execution of the Clear Hip Circle to Handstand (CHCH) on uneven bars based on kinematic parameters. (2) Methods: This study involved elite female gymnasts from the 39th (n = 5, age: 17 ± 6 months) and 40th (n = 8, age: 17.5 ± 6 months) World Cups, totaling 13 gymnasts. Kinematic analysis was performed on 15 successful routines using the Ariel Performance Analysis System (Ariel Dynamics Inc., San Diego, CA). The analysis focused on 16 anthropometric reference points and 8 body segments, including the body mass center of gravity (CG). The main reference points analyzed were the hip joint, the shoulder joint, and the CG along the xy-axes. Trajectory, velocity, angle, and angular velocity of the hips and shoulders were calculated. Pearson correlation analysis was employed to assess the relationships between the kinematic variables. (3) Results: High intercorrelations between the reference points along the xy-axes (0.81–0.99) and optimal movement velocity were found. Dispersed results were observed for kinematic parameters of angle (0.10–0.16) and angular velocity of the hip joints (0.60–1.00), with similar dispersions for shoulder joints (0.51–1.00). Three distinct techniques were identified: (1) stretched body with minimal hip joint flexion throughout; (2) extended body with a short, quick hip joint extension during shoulder movement; and (3) hyperextension in the hip joint. (4) Conclusions: The kinematic analysis revealed three different performance styles of the CHCH among finalists. These variations in technique do not affect the success of the performance. This research contributes to a better understanding of the technique but does not prefer one style over another.

Current state-of-the-art frequency standards are passive optical atomic clocks where the frequency of an optical resonator is stabilized to a narrow atomic transition. Passive clocks have achieved unprecedented stabilities of 6.6 × 10−19 over one hour of averaging time [1]. However, they face intrinsic limitations, particularly due to thermal and mechanical fluctuations of the local oscillator. To surpass the limitations of the passive clocks and go beyond the state-of-the-art, the idea of building active optical atomic clocks emerges. These clocks would be optical counterparts of hydrogen masers, with the emitted frequency defined by the atomic transition and therefore inherently stable against cavity instabilities. This paper discusses the latest developments and future prospects in the field of active optical atomic clocks.

Background Determining human identity has always been important in forensic investigations. Forensic dentistry has developed significantly having a key role in determining gender and age. One of the methods that is important in forensic dentistry is the analysis of orthopantomograms, which are X-rays of the complete upper and lower jaw, including the surrounding anatomical structures. The uniqueness of the dental features recorded in orthopantomograms makes them useful for individual identification, more specifically for the assessment of gender and age. This study was conducted to evaluate the application of convolutional neural networks in automating the process of gender and age estimation based on orthopantomograms, to improve accuracy and efficiency in forensic dentistry. Methodology Convolutional neural networks are powerful tools in the field of artificial intelligence for image processing and analysis because their convolutional layers extract specific features that are characteristic of a certain class. A total of 3716 orthopantomograms collected from the database of the University of Sarajevo - Faculty of Dentistry with the Dental Clinical Center were used to create convolutional neural network models for predicting gender and age. The orthopantomograms were taken in the period from January to December 2022 for the needs of doctors and providing services to patients at four polyclinics: Clinic for Dental Diseases and Endodontics, Clinic for Oral Diseases and Periodontology, Clinic for Oral Surgery, and Clinic for Pediatric and Preventive Dentistry. Results The results derived from three developed models confirm that the developed convolutional neural networks have high accuracy. The first model estimated gender, while the second and the third models estimated age within certain age ranges, the second from 12 to 24 years, and the third from 20 to 70 years. After training on the training dataset, all models achieved high accuracy on the validation dataset. The models demonstrated high accuracy without signs of overfitting, with the first model achieving 95.98%, the second model achieving 97.90%, and the third model achieving 96.12% accuracy. Conclusion This research concluded that the developed convolutional neural networks for gender and age estimation from orthopantomograms showed high accuracy. Models' predictions of gender and two age groups exceeded 95% accuracy. Therefore, convolutional neural networks can be considered useful tools for gender and age determination in forensic dentistry and can facilitate and speed up the processes of assessment and determination of essential characteristics.

AIM To determine whether demographic data, clinical features, and laboratory variables at disease onset can predict the response to methotrexate in juvenile idiopathic arthritis (JIA) patients. METHODS A cohort of 143 newly diagnosed JIA patients initially treated with methotrexate was enrolled in this study. Demographic, clinical, and laboratory parameters were analysed using univariate and multivariate logistic regression to identify predictors of response to methotrexate. The variables included erythrocyte sedimentation rate (ESR), C-reactive protein (CRP), platelets, IgA, IgG, the number of active joints and age at disease onset. Treatment response was assessed at six months, with patients classified as responders (those who achieved clinically inactive disease according to the American College of Rheumatology - ACR criteria) or non-responders. RESULTS Poor response to methotrexate was associated with the number of active joints (p=0.0001; OR=2.7), baseline levels of CRP (p=0.044; OR=1.138), IgA (p=0.004; OR=2.159), and platelet count (p=0.01; OR=1.05). IgG level (P=0.236) did not correlate with the treatment response. CONCLUSION We identified widely available and clinically acceptable biomarkers that can be utilized as predictive indicators of response to methotrexate in JIA patients.

AIM To determine the prevalence of aerobic vaginitis (AV) caused by Enterococcus faecalis (E. faecalis) in human papilloma virus (HPV)-positive women with pathological Pap test and to determine the most prevalent HPV type associated with E. faecalis infection. METHODS This prospective study was conducted at the Gynaecology Centre "Dr. Mahira Jahić" Tuzla and Primary Health Care Centre Tešanj (Bosnia and Herzegovina) in the period between February 2023 and March 2024. The research included 200 women aged 25 to 50 years. The examined group consisted of 100 women with a pathological (examined group) and 100 with a normal (control group) Pap test result. RESULTS Pathological Pap smears were found in 60 (out of 100; 60 %) women in the examined group: cervical intraepithelial neoplasia (CIN) 1 and CIN 2 in two women, respectively, CIN 3 in seven, atypical squamous cells of undetermined significance (ASCUS) in 29 and atypical squamous cells-high-grade cannot be excluded (ASC-H) in two women. Overall (both groups) prevalence of E. faecalis was 25.5% (51women); in 45 (22.5%) women E. faecalis was the only bacterial isolate, of which 42 (21%) in the examined group and three (1.5%) in the control group. High-risk HPV types were found in 62 (out of 100; 62%) women with the pathological Pap smear test. The association of E. faecalis and high-risk HPV positive women was found in 35 (35%) cases (moderately positive correlation; r=0.198). CONCLUSION E. faecalis is very common in HPV 16 and 18 positive women and may represent a risk factor in the development of cervical intraepithelial lesions.

The development of point-of-care wearable devices capable of measuring insulin concentration has the potential to significantly improve diabetes management and life quality of diabetic patients. However, the lack of a suitable point-of-care device for personal use makes regular insulin level measurements challenging, in stark contrast to glucose monitoring. Herein, we report an electrochemical transdermal biosensor that utilizes a high-density polymeric microneedle array (MNA) to detect insulin in interstitial fluid (ISF). The biosensor consists of gold-coated polymeric MNA modified with an insulin-selective aptamer, which was used for extraction and electrochemical quantification of the insulin in ISF. In vitro testing of biosensor, performed in artificial ISF (aISF), showed high selectivity for insulin with a linear response between 0.01 nM and 4 nM (sensitivity of ∼65 Ω nM-1), a range that covers both the physiological and the pathological concentration range. Furthermore, ex vivo extraction and quantification of insulin from mouse skin showed no impact on the biosensor's linear response. As a proof of concept, an MNA-based biosensing platform was utilized for the extraction and quantification of insulin on live mouse skin. In vivo application showed the ability of MNs to reach ISF, extract insulin from ISF, and perform electrochemical measurements sufficient for determining insulin levels in blood and ISF. We believe that our MNA-based biosensing platform based on extraction and quantification of the biomarkers will help move insulin assays from traditional laboratory approaches to personalized point-of-care settings.