Our Work

Preprint. Submitted to Computer Vision and Image Understanding on 5 September 2026; not peer reviewed. The PDF is the submitted manuscript, unchanged.Vision-language models fine-tuned for metric spati…

Preprint. Submitted to Computer Vision and Image Understanding on 5 September 2026; not peer reviewed. The PDF is the submitted manuscript, unchanged.Vision-language models fine-tuned for metric spatial reasoning on one family of sports fail to transfer to another: on basketball imagery, a racket-sport-tuned model underestimates distances as if the 28 m court were ~13 m and reports a ball in nearly every frame. These answer-distribution errors can be repaired without newly authored target QA or answer labels, using the data engine as a verifiable reward oracle: the engine that generates basketball questions from calibrated geometry can also score free-form answers. On an audited protocol with observable referents and instant-disjoint splits, one GRPO pass lifts distance T-MRA from 19.9 to 60.0 and reverses the counting regression. We audit that gain with baselines and controls RLVR studies omit: supervised fine-tuning on the same engine outputs reaches 61.7 at a prompt-matched budget and 71.5-72.4 at token- and update-matched budgets; a per-question constant fitted on the training split, with no image, scores 60.1; the RL policy keeps 40.3 with blanked and 53.5 with mismatched court images, while counting collapses without the image (59.0 vs. 26.0); and option-swapped relational pairs gain no pair-consistent accuracy from the image within noise. The pattern replicates on a second backbone (SmolVLM2-2.2B). Under the tested budgets, rewards, and GRPO implementation, engine-grounded adaptation repairs calibration: the answer distribution moves and metre-scale error shrinks, but 30 cm-level localization does not appear (acc@30cm <= 4.5; 0.0 under a precision-aligned reward), and RL shows no clear distance advantage over supervised consumption of that oracle. We release splits, scorers, predictions, and controls as a benchmark-and-audit package.Code, frozen protocols and evaluation artifacts: doi:10.5281/zenodo.22442707.

Ushbu maqolada yashil iqtisodiyot transformatsiyasi sharoitida barqaror iqtisodiy o‘sish sifatini baholashning nazariy-metodologik asoslarini takomillashtirish masalalari tadqiq etilgan. Tadqiqo…

Ushbu maqolada yashil iqtisodiyot transformatsiyasi sharoitida barqaror iqtisodiy o‘sish sifatini baholashning nazariy-metodologik asoslarini takomillashtirish masalalari tadqiq etilgan. Tadqiqotning asosiy maqsadi iqtisodiy o‘sishning miqdoriy ko‘rsatkichlari bilan bir qatorda uning ekologik barqarorligi, resurs va energiya samaradorligi hamda ijtimoiy natijadorligini kompleks baholashga yo‘naltirilgan metodologik yondashuvni shakllantirishdan iborat. Tadqiqotda tizimli, institutsional va qiyosiy tahlil, statistik guruhlash, indikatorlarni standartlashtirish hamda integral baholash usullaridan foydalanilgan. Natijada iqtisodiy, ekologik va ijtimoiy indikatorlarni o‘zaro integratsiyalash asosida barqaror iqtisodiy o‘sish sifatini baholashning kompleks yondashuvi ishlab chiqilgan. Tadqiqot natijalari yashil transformatsiya sharoitida iqtisodiy o‘sishning samaradorligini baholashda resurslardan oqilona foydalanish, uglerod sig‘imini kamaytirish, qayta tiklanuvchi energiya manbalarini kengaytirish, ekologik investitsiyalar va aholi farovonligi ko‘rsatkichlarini birgalikda hisobga olish zarurligini asoslaydi. Taklif etilgan metodologik yondashuv yashil iqtisodiyotga o‘tish jarayonlarining natijadorligini baholash va barqaror rivojlanish siyosatini ilmiy asoslashga xizmat qiladi.

Цель. Систематизировать современные подходы к лечению больных с острым коронарным синдромом (ОКС) и оценить значение последовательной организации медицинской помощи на догоспитальном, госпитальном и п…

Цель. Систематизировать современные подходы к лечению больных с острым коронарным синдромом (ОКС) и оценить значение последовательной организации медицинской помощи на догоспитальном, госпитальном и постгоспитальном этапах. Материалы и методы. Проведён аналитический обзор современных клинических рекомендаций и результатов крупных рандомизированных исследований, посвящённых диагностике, антитромботической терапии, реперфузионному лечению, чрескожному коронарному вмешательству, вторичной профилактике и кардиореабилитации пациентов с ОКС. Основу анализа составили рекомендации Европейского общества кардиологов 2023 г. и рекомендации ACC/AHA/ACEP/NAEMSP/SCAI 2025 г. Результаты. Современная система лечения ОКС основана на ранней диагностике, быстрой стратификации риска, своевременной реперфузии и индивидуализированной медикаментозной терапии. При STEMI основным методом реперфузии является первичное ЧКВ при возможности его своевременного выполнения. При невозможности своевременного ЧКВ у пациентов без противопоказаний может применяться фибринолитическая терапия с последующей инвазивной стратегией. При NSTE-ACS сроки инвазивного вмешательства определяются клиническим риском. Двойная антитромбоцитарная терапия остаётся основой антитромботической стратегии, однако её продолжительность должна учитывать риск кровотечения. Высокоинтенсивная гиполипидемическая терапия, контроль факторов риска и кардиореабилитация являются обязательными компонентами долгосрочного ведения. Заключение. Эффективность лечения больных с ОКС определяется не отдельным лечебным мероприятием, а согласованностью всей системы медицинской помощи. Сокращение времени до реперфузии, правильная маршрутизация, применение доказанных методов реваскуляризации, индивидуализация антитромботической терапии и проведение вторичной профилактики позволяют сформировать непрерывную модель лечения пациента.

The Open Radio Access Network (O-RAN) paradigm is a promising network architecture that promotes flex ibility and openness towards formulating interoperable, software driven and vendor-agnostic 6G net…

The Open Radio Access Network (O-RAN) paradigm is a promising network architecture that promotes flex ibility and openness towards formulating interoperable, software driven and vendor-agnostic 6G networks. However, to realize its remarkable capabilities, efficient resource management is critical. In this work, we examine the O-RAN service provisioning capability under the probabilistic Bandwidth Reservation (BR) policy and we benchmark its performance against two classical resource management policies, i.e. the Complete Sharing (CS) and the strict BR. For our analysis, we propose a novel multi-rate loss model that incorporates the concept of the probabilistic BR. Numerical results demonstrate that the adopted resource man agement policy highly affects the user service, with probabilistic BR offering the best trade-off by maintaining prioritization for high-priority applications and in parallel limiting the service degradation experienced by others.

This dataset contains supplementary tables for the manuscript titled "Thermodynamic Quantification of the H2-generation Potential of Earth's Igneous Rocks for Geologic H2 Exploration". Downloadable su…

This dataset contains supplementary tables for the manuscript titled "Thermodynamic Quantification of the H2-generation Potential of Earth's Igneous Rocks for Geologic H2 Exploration". Downloadable supplementary data includes: Table S1 – Composition of reacting fluids Table S2 – Molar Fe(III)/FeT assigned to each rock type and their data sources Table S3 – List of rock types used in this work and the number (n) of samples for each rock type used in the simulations Table S4 – Rock compositions used in simulations and calculated H2-generation potentials at different model scenarios Table S5 – List of minerals included in simulations Table S6 – List of mineral solid solutions used in simulations.

OverviewThis repository provides the complete replication materials, agent-based simulation codebase (Mesa 3.0), empirical datasets, cyber-physical edge perception prototypes (YOLOv8 + OpenCV), and hi…

OverviewThis repository provides the complete replication materials, agent-based simulation codebase (Mesa 3.0), empirical datasets, cyber-physical edge perception prototypes (YOLOv8 + OpenCV), and high-resolution publication figures for the research paper: Title: Trustworthy AI for Sustainable Cities: A Multimodal Orchestration Agent-Based Framework for Civic Compliance • License: MIT Open-Source License Key Replication Artifacts Included: 1. Multi-Agent Behavioral Simulation Engine (Mesa 3.0): 100-Run Monte Carlo Simulation: Automated trajectory generation over 80 simulation steps (run_monte_carlo.py), reproducing primary empirical findings: HARMONY terminal compliance rate (87.90% ± 3.30% [95% CI: 87.25%, 88.55%]) vs. the coercive enforcement baseline (57.00% ± 4.89% [95% CI: 56.04%, 57.96%]), demonstrating a net divergence of +30.90% (Section 3.1 & Figure 5). Six-Arm Factorial Ablation Matrix: Full quantitative decomposition (run_factorial_ablation.py) across all six experimental arms and five civic population cohorts (Champions, Ambivalent, Resistant, Heterogeneous, and Stress-Test), directly validating Table 7 and Table 8. Multivariate Sensitivity Sweeps: Parameter resilience analysis across politeness modifiers (L_p ∈ [0.0, 1.0]), gratitude reinforcement (δ ∈ [0.005, 0.050]), acoustic noise (50–100 dB), and the irreducible defiance ceiling (P_comp(max) = 1 - ε = 88.0%; Section 3.3 & Figure 8). 2. Master Cyber-Physical Edge Perception Prototype (YOLOv8 + OpenCV): Live Webcam Prototype (src/edge_perception/harmony_master.py): Real-time spatial perception, dynamic bounding-box tracking, contextual auditory nudging (nudge.mp3), gratitude reinforcement (thankyou.mp3), and a publication-grade two-tier HUD with dual-rate confidence accumulation (+2 credits/frame for direct bin aperture; +1 credit/frame for peripheral pickup, capping at τ_verify = 30 credits). Headless Reviewer Benchmark: Parameterized profiling benchmark (--benchmark --frames 600) validating real-time embedded timing performance (mean latency: 18.3 ms ± 1.1 ms, 54.6 FPS). 3. Empirical CSV Datasets (data/): Harmony_results_harmonized.csv: Full 100-run trajectory time-series across 80 epochs. Ablation_study_results.csv: Six-arm factorial ablation and cohort compliance metrics. Sensing_resilience_results.csv: Acoustic interference sweep data across 50–100 dB. 4. Publication Graphics & Vector Figures (figures/): Complete suite of 300+ DPI raster images and vector PDFs for Figures 1 through 9, including the Four-Phase Civic Interaction Lifecycle diagram. 5. Complete Bibliographic Libraries (references/): Mendeley/Zotero-compatible BibTeX (.bib) and EndNote/Zotero-compatible RIS (.ris) libraries with active DOIs for all 51 references cited in the manuscrip Deterministic Execution GuaranteeAll stochastic routines are governed by deterministic pseudo-random seed sequencing (Seed_k = base_seed + k, where base_seed = 42). Reviewers and researchers can reproduce all empirical figures, statistical dispersion metrics (±1σ SD), and inferential precision intervals (95% bootstrap CI) with a single command:py run_all_validation.py Ethical Governance & Privacy PreservationOperates under an Edge-Only Ephemeral Vector Processing (E-EVP) paradigm compliant with GDPR Article 25 (Data Protection by Design), NIST AI RMF 1.0 (Govern, Map, Measure, Manage), and ISO/IEC 23894:2023. Video frames are processed in volatile SRAM and discarded within 200 ms; zero biometric identifiers or persistent video recordings are stored or transmitted.

Miasteniya gravis (MG) nerv-mushak uzatilishining autoimmun kasalligi bo'lib, uning kechishida immun, metabolik va nutritiv omillarning o'zaro ta'siriga qiziqish ortmoqda. Ushbu narrativ tanqidiy shar…

Miasteniya gravis (MG) nerv-mushak uzatilishining autoimmun kasalligi bo'lib, uning kechishida immun, metabolik va nutritiv omillarning o'zaro ta'siriga qiziqish ortmoqda. Ushbu narrativ tanqidiy sharh MGda mikronutrient va mikroelement o'zgarishlariga oid klinik dalillarni umumlashtiradi, bevosita inson ma'lumotlarini biologik gipotezalardan ajratadi va tadqiqot bo'shliqlarini belgilaydi. Vitamin D eng ko'p o'rganilgan bo'lib, ayrim kohortlarda uning past darajasi qayd etilgan, ammo sababiylik tasdiqlanmagan. Temirga oid ma'lumotlar oddiy tanqislikdan ko'ra gomeostaz o'zgarishini ko'rsatadi. Tarixiy rux tadqiqotlari ayrim MG guruhlarida qondagi rux oshishi va timus-immun faolligi bilan bog'liqlikni ko'rsatgan. Kalsiy va magniy asosan nerv-mushak fiziologiyasi va davolash xavfsizligi nuqtai nazaridan muhim. Mavjud dalillar MG uchun universal mikronutrient yetishmovchiligi modelini tasdiqlamaydi; istiqboldagi ishlar mikronutrient profillarini yallig'lanish, timus fenotipi, autoantitana holati, davolash ekspozitsiyasi va standart klinik natijalar bilan birga baholashi kerak.

Evidence for large-scale differential rotation in the early universe and forward generation of its CMB topology by a primordial magnetic source — v1.8 reproducibility release Canonical 3° AC…

Evidence for large-scale differential rotation in the early universe and forward generation of its CMB topology by a primordial magnetic source — v1.8 reproducibility release Canonical 3° AC* is applied to the Planck PR3 SMICA, NILC, Commander, and SEVEM temperature reconstructions. AC* is a support-indexed multiscale spatial reconstructibility operator; direction, topology, and physical rotation arise only downstream. For canonical SMICA, 36,459 valid supports yield 27 large semiarcs. A separate high-resolution route uses 151 reciprocal HR1–HR2 structures to reconstruct an axis-free spherical tangent field. Blind recovery finds exactly four critical points of the represented lmax = 4 field: two spiral foci, one Node, and one Saddle. An all-sphere interval/Krawczyk analysis establishes completeness of this zero set. Focus A and Focus B are distinct oppositely rotating structures and admit a common transverse regular-vector realization in the coupled Einstein–Boltzmann state at recombination. Together with the differential-rotation reconstruction, polarization consistency, four-map comparison, perturbation tests, robustness controls, and exact propagation to z = 0, this provides evidence for large-scale differential rotation in the early universe. The independent forward-source test propagates a near-scale-invariant primordial magnetic source with nB = −2.9 to synthetic CMB skies and applies the same reconstruction without using the observed temperature map, critical-point geometry, Focus-A/Focus-B positions, measured H(J), or a magnetic-amplitude fit. One coordinate-blind realization generates the measured 2F+N+S topology and both rotational foci. The B-associated synthetic focus differs from the observed B focus only in the sign of a small divergent component; exact critical-point geometry is not reproduced. In eight matched phase-realization pairs, complete reconstruction occurs in five near-scale magnetic cases and none of the causal-white comparator cases, with three ties. The result establishes forward-generative capacity of the tested primordial magnetic source for the measured CMB topology and its two-focus rotational structure. It does not determine a unique magnetic spectrum, exclusivity among all possible source mechanisms, or an ensemble occurrence rate, and it is not a direct magnetic-helicity test. The release contains four public objects:Rambold_2026_CMB_v1p8.pdf — scientific manuscript.CMB_v1p8_interactive_scientific_viewer.zip — interactive scientific viewer.CMB_v1p8_python_pipeline.zip — executable analysis and replay code.CMB_v1p8_reproducibility_records.zip — scientific products, replay evidence, source-generation records, provenance, licences, manifests, SHA-256 bindings, and integrity records. Planck and WMAP FITS products are not redistributed and remain available from their public archives. Author-created scientific text and derived materials are released under CC BY 4.0 where applicable; original author-created software is released under the MIT License unless stated otherwise.