Our Work

Deep-learning background field removal (BFRnet): a 3D dual-frequency octave-convolution U-net trained to predict the background field of the brain — including brains with significant pathological susc…

Deep-learning background field removal (BFRnet): a 3D dual-frequency octave-convolution U-net trained to predict the background field of the brain — including brains with significant pathological susceptibility sources (haemorrhage, calcification). Consumes the total field (ppm) and predicts the background field; the local tissue field is total − background, masked. The authors' trained MATLAB network was exported to ONNX and is run here with ONNX Runtime — no MATLAB Runtime — reproducing the MATLAB output to ~1e-8 at a fraction of the memory and image size. QSM-CI reconstruction method bfrnet. Browse and run it at https://qsmxt.github.io/QSM-CI/submission.html?method=bfrnet.

AutoQSM — learning-based single-step QSM reconstruction without brain extraction. A V-Net that maps the total field map directly to susceptibility, doing neither brain extraction nor a separate backgr…

AutoQSM — learning-based single-step QSM reconstruction without brain extraction. A V-Net that maps the total field map directly to susceptibility, doing neither brain extraction nor a separate background-field-removal step. Consumes the total field and produces susceptibility. Runs CPU-only on the legacy TensorFlow 1.15 / Keras 2.2.5 stack the method was published against. QSM-CI reconstruction method autoqsm. Browse and run it at https://qsmxt.github.io/QSM-CI/submission.html?method=autoqsm.

Approximate Message Passing with Parameter Estimation: a probabilistic Bayesian dipole inversion. MAP estimation over the nonlinear complex-exponential forward model with a Laplace sparse-wavelet prio…

Approximate Message Passing with Parameter Estimation: a probabilistic Bayesian dipole inversion. MAP estimation over the nonlinear complex-exponential forward model with a Laplace sparse-wavelet prior and a two-component Gaussian-mixture noise model, with all distribution parameters estimated automatically (tuning-free). Compiled from MATLAB and run on the license-free MATLAB Runtime. QSM-CI reconstruction method amp-pe. Browse and run it at https://qsmxt.github.io/QSM-CI/submission.html?method=amp-pe.

This article investigates the tribotechnical properties of heterocomposite polymer materials created using local industrial waste based on epoxy-phenol-formaldehyde hybrid reactoplastic binders. The i…

This article investigates the tribotechnical properties of heterocomposite polymer materials created using local industrial waste based on epoxy-phenol-formaldehyde hybrid reactoplastic binders. The influence of kaolin, graphite, pyrolysis residues, and short-staple silk waste on the properties of the composite material was evaluated. Based on FE-SEM and X-ray spectral analyses, the microstructure and elemental composition of the materials were studied, and the possibilities of increasing the thermal stability, interphase synergetic effect, and wear resistance of hybrid matrices were scientifically substantiated.

Аннотация. В статье анализируется влияние моделей кредитного скоринга на основе больших данных и машинного обучения на выручку финтех-компаний. Автор рассматривает механизмы роста дохода через уровень…

Аннотация. В статье анализируется влияние моделей кредитного скоринга на основе больших данных и машинного обучения на выручку финтех-компаний. Автор рассматривает механизмы роста дохода через уровень одобрения, уровень дефолта и операционные издержки, а также сопоставляет типы скоринговых моделей по их экономической отдаче. Ключевые слова: кредитный скоринг, большие данные, машинное обучение, финтех, рост выручки, уровень дефолта, уровень одобрения, альтернативные данные, финансовая инклюзия, кредитный риск, операционные издержки. Annotatsiya. Maqolada katta ma’lumotlar va mashinali o‘qitish asosidagi kredit skoringi modellarining fintex kompaniyalari daromadiga ta’siri tahlil qilinadi. Muallif tasdiqlash darajasi, defolt darajasi va operatsion xarajatlar orqali daromad o‘sishi mexanizmlarini ko‘rib chiqadi hamda skoring modellari turlarini qiyoslaydi. Kalit so‘zlar: kredit skoringi, katta ma’lumotlar, mashinali o‘qitish, fintex, daromad o‘sishi, defolt darajasi, tasdiqlash darajasi, muqobil ma’lumotlar, moliyaviy inklyuziya, kredit riski, operatsion xarajatlar. Abstract. This article analyzes the impact of Big Data- and machine-learning-based credit scoring models on the revenue of fintech companies. The author examines revenue-growth mechanisms through approval rates, default rates and operating costs, and compares scoring model types by their economic returns. Keywords: credit scoring, big data, machine learning, fintech, revenue growth, default rate, approval rate, alternative data, financial inclusion, credit risk, operating costs.

This paper presents MyEvac, a real-time flood evacuation  alert architecture that integrates citizen-submitted images and videos, analysed by Qwen2.5-VL-72B, into a three-source weighted confiden…

This paper presents MyEvac, a real-time flood evacuation  alert architecture that integrates citizen-submitted images and videos, analysed by Qwen2.5-VL-72B, into a three-source weighted confidence engine alongside official JPS river sensor data and Open-Meteo GloFAS v4 probabilistic forecasts. The system operationalises a distinctive integration pattern: VLM-derived structured flood variables - water depth, severity classification, visual indicators, Malay OCR-derived location, and temporal water rise rate - enter the alert confidence formula as a calibrated weighted term, controlled by an explicit two-of-three source consensus gate before geofenced alert dispatch. The system was verified through 150 automated tests and initially validated through four real-world Malaysian flood media inputs and live external API integration. Prototype deployment targets the Taman Sri Muda flood zone in Shah Alam, Selangor, Malaysia.

This article analyzes the emergence and development stages of the automatic transmission, the operational characteristics of mechanical transmission synchronizers, and their impact on driver performan…

This article analyzes the emergence and development stages of the automatic transmission, the operational characteristics of mechanical transmission synchronizers, and their impact on driver performance. The first automatic transmission developed by General Motors and its evolution from a hydraulic clutch to a hydraulic transformer are highlighted. Additionally, the structural design of the hydraulic transformer, the principle of torque transmission, and the advantages of automatic torque control in accordance with the vehicle's driving conditions have been scientifically substantiated.  

Snapshot of data of the Tool Registry for Digital Humanities from Wikidata. Data is being downloaded through a combination of SPARQL queries and API calls and serialised as Turtle and JSON-LD.

LLM-based agents are often criticized for lacking spatial understanding and mainly exploiting statistical text patterns. We investigate their spatial comprehension through an architecture combining ge…

LLM-based agents are often criticized for lacking spatial understanding and mainly exploiting statistical text patterns. We investigate their spatial comprehension through an architecture combining geometrical tools with a large language model serving as a high-level orchestrator in grid-world environments. The agent first collects geodesic trajectories, which are then vector‑quantized to extract a representative subset. Each resulting trajectory defines a reusable tool. The LLM is used offline to interpret the learned skills by projecting them into natural language descriptions of the underlying behavioral patterns. Online, the LLM chooses the appropriate tool conditioned on the current state and goal. Low-level control is handled by primitive actions that execute the selected skill. From an agentic AI perspective, this approach separates learning into two levels. Skill discovery is handled through unsupervised quantization of trajectories, while reasoning and decision-making are handled by the LLM.