Our Work

JAX reimplementation and bindings for NVIDIA's cuDecomp library, enabling multi-node parallel FFTs and halo exchanges directly in low-level NCCL/CUDA-Aware MPI from JAX code. jaxDecomp provides both a…

JAX reimplementation and bindings for NVIDIA's cuDecomp library, enabling multi-node parallel FFTs and halo exchanges directly in low-level NCCL/CUDA-Aware MPI from JAX code. jaxDecomp provides both a pure JAX backend for easy deployment and an optional cuDecomp backend for advanced MPI/NCCL communication patterns. All functions are JIT-compatible and support automatic differentiation.

Despite more than a decade of machine learning (ML) research on Alzheimer's disease (AD), early detection remains largely unsolved in practice: high-accuracy models are concentrated in neuroimaging-ba…

Despite more than a decade of machine learning (ML) research on Alzheimer's disease (AD), early detection remains largely unsolved in practice: high-accuracy models are concentrated in neuroimaging-based deep learning, which is costly and inaccessible in resource-constrained settings, while studies using cheaper, structured clinical data rarely combine rigorous preprocessing, systematic hyperparameter optimisation, class-imbalance handling, and a validated deployment pathway within a single, reproducible framework. This paper addresses that gap. It proposes and experimentally validates a comparative ML framework for early AD detection using structured clinical, demographic, lifestyle, and cognitive assessment data. A publicly available Kaggle dataset comprising 2,149 patient records and 35 attributes was used. Three supervised classifiers—Support Vector Machine (SVM), Random Forest (RF), and Logistic Regression (LR) were developed within a reproducible preprocessing pipeline, optimised using GridSearchCV with stratified cross-validation, and evaluated under two experimental conditions: with and without Synthetic Minority Oversampling Technique (SMOTE). The RF classifier achieved the best performance on the held-out test set, with 94.42% accuracy, 91.95% F1-score and 93.93% ROC-AUC, and it further exhibited strong robustness to class imbalance, with only marginal performance variation across the SMOTE and non-SMOTE conditions. In contrast, LR benefited substantially from SMOTE (F1-score improving from 73.58% to 76.70%), while SVM performance declined slightly. Compared against a closely related baseline study evaluating SVM, RF, and CNN on a similarly structured Kaggle dataset, the proposed RF model outperformed the baseline's traditional ML results while remaining competitive with its CNN-based results, without requiring neuroimaging data. The validated model was further deployed as an interactive Streamlit-based prediction application, with an embedded Developer Mode empirically confirming prediction consistency between offline experimentation and real-time deployment. These findings demonstrate that a carefully optimised, ensemble-based ML framework operating on non-invasive structured data can achieve potentially useful screening performance while offering limited global interpretability through feature-importance estimates, being computationally lightweight, and deployable, making it well-suited to resource-constrained healthcare screening contexts such as Malaysia.

https://pvlib-python.readthedocs.io/en/v0.16.1/whatsnew.html

ОШҚОЗОН ВА ЎН ИККИ БАРМОҚЛИ ИЧАК ЯРАСИ АСОРАТЛАРИ – ҚОН КЕТИШЛАРДА ЖАРРОҲЛИК ТАКТИКАСИНИ ТАНЛАШНИ ОПТИМАЛЛАШТИРИШ. ² Mavlonov Abduqahhor Xudoyberdiyevich, ¹ Xursanov Yoqubjon Erkin o&l…

ОШҚОЗОН ВА ЎН ИККИ БАРМОҚЛИ ИЧАК ЯРАСИ АСОРАТЛАРИ – ҚОН КЕТИШЛАРДА ЖАРРОҲЛИК ТАКТИКАСИНИ ТАНЛАШНИ ОПТИМАЛЛАШТИРИШ. ² Mavlonov Abduqahhor Xudoyberdiyevich, ¹ Xursanov Yoqubjon Erkin o‘g‘li, ³ Xidirov Muzaffar Sultonazarovich, ⁴ Mirzaboyev Abdurazzoq Mamirovich, ⁵ Karayev Muzrob Nuraliyevich ¹ Samarqand davlat tibbiyot universiteti 2-son xirurgik kasalliklar kafedrasi dotsenti, ² Toyloq tumani tibbiyot birlashmasi markaziy shifoxonasi bosh shifokori, ³ Respublika shoshilinch tibbiy yordam ilmiy markazi Samarqand filiali shifokori, ⁴ Respublika shoshilinch tibbiy yordam ilmiy markazi Samarqand filiali shifokori, ⁵ Respublika shoshilinch tibbiy yordam ilmiy markazi Samarqand filiali shifokori. https://doi.org/10.5281/zenodo.22975197 Аннотация: Ушбу илмий ишда гастродуоденал ярадан қон кетиш (ГДЯК) ҳолатларини жарроҳлик йўли билан даволашдаги замонавий тактик ва техник ёндашувларнинг аҳамияти ва самарадорлиги таҳлил қилинган. ГДЯК қорин бўшлиғи шошилинч жарроҳлигидаги жиддий муаммолардан бири бўлиб, беморларнинг ўлими юқори даражада сақланиб қолмоқда. Тахминан 20–25% ҳолларда бирламчи эндоскопик усуллар билан қон кетишни тўхтатишга эришилмайди ва жарроҳлик аралашуви талаб этилади.Ваготомия каби органни сақлайдиган усуллар ГДК (гастродуоденал қон кетишлар)ни самарали даволашда асосий ўринни эгаллайди. Эндоскопик ва мини-инвазив аралашувларнинг клиник амалиётда кенг қўлланиши жарроҳлик кўрсатмаларини аниқлашда муҳим аҳамият касб этади.Тадқиқот натижаларига кўра, замонавий технологик ёндашувлар ва органни сақловчи операциялар қон кетишни самарали тўхтатиш, асоратларни камайтириш ва беморларнинг яшаш прогнозини яхшилашда муҳим омил ҳисобланади. Калит сўзлар: гастродуоденал ярали қон кетиш; эндоскопик гемостаз; гибрид жарроҳлик; лапароскопия; Форрест таснифи; қон кетишнинг қайталаниши.

Brain tumor is among the most life-threatening neurological disorders, arising from the uncontrolled proliferation of abnormal cells within the brain and its surrounding tissue. Precise and early iden…

Brain tumor is among the most life-threatening neurological disorders, arising from the uncontrolled proliferation of abnormal cells within the brain and its surrounding tissue. Precise and early identification of the tumor region is essential for timely clinical intervention, yet single-modality imaging such as Magnetic Resonance Imaging (MRI) alone often struggles to capture both the structural and the metabolic characteristics of a lesion. Positron Emission Tomography (PET) provides complementary functional information that, when fused with MRI, improves the reliability of tumor localization. However, existing multimodal frameworks frequently lose high-frequency structural detail during fusion, align cross-modal features poorly, and depend on manually tuned hyperparameters that limit generalization across patients. To address these shortcomings, this study designs a Multihead Cross-Covariance Attention Fusion-based Adaptive Explainable Pyramid Dilated Res-DenseNet (MCCAF-AExPDRDNet) for multimodal brain tumor segmentation from paired MRI and PET slices. The proposed dual-branch network extracts structural and metabolic features independently through pyramid dilated dense blocks and residual dense blocks, and fuses the resulting bottleneck representations through a multihead cross-covariance attention mechanism before reconstructing the tumor mask through a skip-connected decoder. To remove the burden of manual hyperparameter selection, a Best-fit Guided Learning based Apiary Organizational-based Optimization algorithm (BGL-AOO) is proposed to adaptively tune the hidden neuron count, learning rate, and steps per epoch of the segmentation network. The framework is validated on a real, paired MRI-PET brain dataset against U-Net, ResUNet, TransUNet, GARU-Net, MBTC-Net and ResNet18 baselines, and against Educational Competition Optimization (ECO), Quokka Swarm Optimization (QSO), the Supercell Thunderstorm Algorithm (STA) and standard Apiary Organizational-based Optimization (AOO). Across six activation functions, the proposed BGL-AOO-MCAF-XPDN model attains a mean accuracy of up to 95.58%, a Dice coefficient of up to 95.60%, an IoU of up to 91.57%, a recall of up to 95.66%, and a PSNR of up to 61.70 dB, consistently outperforming every comparator model and optimizer. These outcomes indicate that the designed framework offers a scalable and clinically relevant tool for computer-aided brain tumor diagnosis.

As the number of tasks and virtual machines increases for task scheduling in cloud computing environments, predicting an efficient load balancing model for multi-server, multi-task scheduling becomes …

As the number of tasks and virtual machines increases for task scheduling in cloud computing environments, predicting an efficient load balancing model for multi-server, multi-task scheduling becomes challenging. Many conventional models rely on static multi-task scheduling algorithms for decision-making. Additionally, these models utilize traditional classification algorithms for generating multi-task patterns, often working with limited data sizes. This work presents a hybrid load balancing framework designed to enhance decision pattern mining in the multi-server, multi-task scheduling process. The framework integrates Particle Swarm Optimization (PSO) and Ant Colony Optimization (ACO) into the ensemble load balancing approach within a real-time cloud computing environment. Furthermore, a hybrid decision tree classifier is proposed to facilitate decision-making in the multi-task scheduling process. Experimental results demonstrate that the proposed model outperforms conventional models in terms of multi-task, multi-server scheduling in real-time cloud computing environments. 

Code and derived data supporting the manuscript "Origin of the conflicting hole masses in the GaN/AlN two-dimensional hole gas" by T. M. Mahim, A. S. M. Mohsin and M. M. Rahman. This deposit does not …

Code and derived data supporting the manuscript "Origin of the conflicting hole masses in the GaN/AlN two-dimensional hole gas" by T. M. Mahim, A. S. M. Mohsin and M. M. Rahman. This deposit does not contain the manuscript. Source code: https://github.com/Tanvir-Mahmud-Mahim/gan-2dhg-masses-lifetimes THIS VERSION (v8) accompanies a further revision of the manuscript. It adds: (i) data/chang2026_fig2_digitized.json replaced by a reading of Fig. 2 of Chang et al. (arXiv:2501.16213v1) at the native resolution of the embedded image (2498 x 1677 pixels, 0.24 ohm per pixel, ten temperatures), which returns their light-hole Dingle mobility (352 to 385 cm2/Vs at 1.8 to 6.0 K, reported 368 +/- 14) and heavy-hole mass (1.97 +/- 0.13 m0, reported 1.92 +/- 0.16); (ii) src/gan2dhg/sdh.py with the density-of-states factor in sigma_xx (Dmitriev et al., Rev. Mod. Phys. 84, 1709 (2012)), the spin factor from two harmonics and a non-perturbative rho_xx with Lorentzian Landau levels to all harmonics; (iii) scripts/run_heavy_quantum_mobility.py (a joint Dingle fit of the heavy-hole oscillation, 74 cm2/Vs raw, and the amplitude-ratio route with the light-hole spin factor 0.745 measured from the second harmonic, 95 cm2/Vs) and scripts/run_forward_calibration.py (calibration of both routes, 94 and 96 cm2/Vs); the adopted heavy-hole quantum mobility is 95 cm2/Vs instead of 167 to 200; (iv) scripts/run_revised_mobilities.py and run_light_corrected.py: with interface roughness, a long-range component with |V(q)|^2 ~ q^-p fits the four mobilities within 2 percent with p = 3.55 at 95 cm2/Vs (within 4 percent with p = 3.2 when the light-hole quantum mobility is corrected in the same way); kernels for charged lines in the interface and a power-law spectrum in scatter2d.py; (v) scripts/disorder_fit.py (shared fitting). The test suite has 61 tests, all passing. EARLIER VERSIONS (v1 to v7) are described on their own records. Where their descriptions give a different heavy-hole quantum mobility or a different account of the disorder, the values of this version supersede them. CONTENTS: src/gan2dhg (six-band Hamiltonian, hard-wall and finite-barrier self-consistent solvers, Landau levels, two-dimensional scattering, oscillation amplitudes in rho_xx, published values); scripts (one script per result file, and the figure scripts); results (JSON outputs, including cached Landau levels); data/gan_2dhg_measured.yaml (provenance of every published value and the citation audit); data/chang2026_fig2_digitized.json (Fig. 2 of Chang et al. read at native resolution, with the calibration); tests. No experiment was performed for this work; every experimental number is a published value, recorded with its source.

Background: Football trafficking is a growing form of exploitation facing young Nigerian players, who are deceived by fraudulent agents promising trials, contracts and international mobility. Understa…

Background: Football trafficking is a growing form of exploitation facing young Nigerian players, who are deceived by fraudulent agents promising trials, contracts and international mobility. Understanding victims’ lived experiences alongside the communication interventions of media professionals, advocacy groups, academy coaches, and football organisations is essential for effective prevention.  Objective: This study examined the lived experiences of football trafficking victims in Nigeria and the communication interventions deployed by stakeholders, together with the constraints that limit their implementation and sustainability.  Method: A descriptive phenomenological design was employed. Twenty-seven participants (21 male, 6 female) were purposively sampled: five victims, three organisational spokespersons, nine media professionals, five academy coaches and five advocacy members. In-depth interviews were analysed thematically using Braun and Clarke’s (2006) six-phase framework.  Results: Poverty, limited education and aspirations for socioeconomic mobility increased vulnerability to deceptive recruitment, resulting in financial loss, exploitative contracts, poor living conditions and psychological trauma. Stakeholders used community outreach, seminars, workshops, interpersonal communication, social media and educational programmes, yet inadequate funding, personnel shortages, logistical barriers, low public awareness and weak inter-agency collaboration undermined effectiveness.  Conclusion: Football trafficking remains a serious threat to economically disadvantaged Nigerian youth. Existing communication interventions, though well-intentioned, are fragmented, under-resourced and insufficiently coordinated to counter sophisticated trafficking networks.  Unique Contribution: The study is the first in the Nigerian context to connect victims’ first-hand accounts with stakeholders’ reported interventions and constraints, bridging experiential and institutional perspectives.  Key Recommendation: Stakeholders should develop a coordinated, adequately funded, multi-agency communication strategy, anchored in schools, academies and digital platforms, that directly addresses the identified resource and coordination gaps. 

Although the resource-based view predicts that managerial innovation should enhance firm performance, empirical evidence remains contradictory and is drawn overwhelmingly from large firms in developed…

Although the resource-based view predicts that managerial innovation should enhance firm performance, empirical evidence remains contradictory and is drawn overwhelmingly from large firms in developed economies, leaving the financial payoff of managerial innovation for small and medium-sized enterprises (SMEs) in emerging markets such as Morocco poorly understood. The primary objective of this research is to investigate the impact of managerial innovation and digital technology on the financial performance of SMEs from 2020 to 2023, aiming to construct a model that elucidates the influence of managerial innovation dimensions, with particular attention to the technological dimension, on financial performance. The research framework developed in this study was tested on 260 Moroccan SMEs operating across various sectors between 2020 and 2023. Adopting a positivist epistemological stance, the study employed a hypothetico-deductive reasoning approach and a quantitative research design. The empirical validation of the proposed model provided evidence of a partially significant relationship between managerial innovation dimensions and firm financial performance. The results revealed that strategic planning, employee development and motivation, inter-organizational relations and partnerships, and information and communication technologies have a statistically significant impact on financial performance. In contrast, the structural dimension of managerial innovation did not exhibit a statistically significant relationship with financial performance. This finding suggests that structural changes may not necessarily translate into immediate financial benefits for SMEs, particularly in contexts characterized by economic uncertainty and resource constraints. This study enables SME managers to identify the organizational changes required to foster managerial innovation and improve performance. It also contributes to the existing literature by providing empirical evidence on managerial innovation within the context of Moroccan SMEs. By identifying which dimensions of managerial innovation reliably translate into financial gains and which do not, the study offers SME managers and policymakers an evidence-based basis for prioritizing investment under conditions of resource scarcity.