arXiv:2508.13378v3 Announce Type: replace 
Abstract: Small Language Models (SLMs) are a practical option for narrow, workflow-relevant medical imaging utilities where privacy, latency, and cost dominate. We present a governance-ready recipe that combines prompt scaffolds, calibrated abstention, and standards-compliant integration into Picture Archiving and Communication Systems (PACS). Our focus is the assistive task of AP/PA view tagging for chest radiographs. Using four deployable SLMs (Qwen2.5-VL, MiniCPM-V, Gemma 7B, LLaVA 7B) on NIH Chest X-ray, we provide illustrative evidence: reflection-oriented prompts benefit lighter models, whereas stronger baselines are less sensitive. Beyond accuracy, we operationalize abstention, expected calibration error, and oversight burden, and we map outputs to DICOM tags, HL7 v2 messages, and FHIR ImagingStudy. The contribution is a prompt-first deployment framework, an operations playbook for calibration, logging, and change management, and a clear pathway from pilot utilities to reader studies without over-claiming clinical validation. We additionally specify a human-factors RACI, stratified calibration for dataset shift, and an auditable evidence pack to support local governance reviews.

تظهر النماذج اللغوية الصغيرة (SLMs) كأدوات فعالة لمهام محددة في التصوير الطبي، خاصة في البيئات التي تهيمن عليها الخصوصية والكمون والتكلفة. يقدم هذا المقال إطار عمل جاهز للحكومة يجمع بين هياكل المطالبات، والامتناع المعاير، والتكامل المتوافق مع أنظمة أرشفة الصور والتواصل (PACS). يركز البحث على وسم عرض AP/PA للأشعة السينية للصدر، باستخدام أربعة نماذج لغوية صغيرة قابلة للنشر. تشير النتائج إلى أن المطالبات الموجهة للتفكير تفيد النماذج الأخف، بينما تكون النماذج الأقوى أقل حساسية لهذه المطالبات.

Los pequeños modelos de lenguaje (SLMs) están surgiendo como herramientas efectivas para tareas específicas de imagenología médica, especialmente en entornos donde la privacidad, la latencia y el costo son críticos. Este artículo presenta un marco listo para la gobernanza que integra andamiajes de prompts, abstención calibrada y cumplimiento con los sistemas de archivo y comunicación de imágenes (PACS). El enfoque está en el etiquetado de vistas AP/PA para radiografías de tórax, utilizando cuatro SLM desplegables. Los hallazgos indican que los prompts orientados a la reflexión mejoran el rendi…

Les petits modèles de langage (SLMs) émergent comme des outils efficaces pour des tâches spécifiques d'imagerie médicale, en particulier dans des environnements où la confidentialité, la latence et le coût sont critiques. Cet article présente un cadre prêt à la gouvernance qui intègre des échafaudages de prompts, une abstention calibrée et une conformité avec les systèmes d'archivage et de communication d'images (PACS). L'accent est mis sur le marquage de vues AP/PA pour les radiographies thoraciques, en utilisant quatre SLM déployables. Les résultats indiquent que les prompts orientés vers la…

Small Language Models (SLMs) are emerging as effective tools for specific medical imaging tasks, particularly in environments where privacy, latency, and cost are critical. This article presents a governance-ready framework that integrates prompt scaffolds, calibrated abstention, and compliance with Picture Archiving and Communication Systems (PACS). The focus is on AP/PA view tagging for chest radiographs, utilizing four deployable SLMs. The findings indicate that reflection-oriented prompts enhance performance in lighter models, while stronger models show less sensitivity to these prompts.

Governance-Ready Small Language Models for Medical Imaging: Prompting, Abstention, and PACS Integration

Walmart's early Black Friday deals are here, with big discounts on TVs, laptops, and more (including the lowest-ever prices on AirTags, AirPods, and the Apple Watch Series 10).

Best Walmart Black Friday deals live now: Save up to 60% on Apple, Dyson, TVs, and more

Matter 1.5 may finally fix the biggest headache in buying security cameras - here's how

Organize your workspace with our top picks for the best laptop docking stations available now.

I tested the best laptop docking stations - here's what I recommend for your office setup

Yann LeCun, an artificial intelligence pioneer who runs a research lab at Meta Platforms Inc., told employees that he will depart the social media giant at the end of the year and start a new company, according to a memo obtained by Bloomberg News.

Meta AI pioneer LeCun announces exit, plans new startup

In late September, the German airline Deutsche Lufthansa AG told analysts and investors that it planned to eliminate 4,000 administrative positions by the end of the decade. Among the reasons it cited was "the increased use of artificial intelligence."

Companies are warming up to saying AI is the reason for job cuts

The Xgimi Horizon S Max stands out as one of the brightest and most capable projectors I've tested to date.

Finally, a portable 4K projector worthy of replacing my TV - and it supports Dolby Vision

arXiv:2511.15661v1 Announce Type: cross 
Abstract: Reinforcement learning (RL) provides a principled framework for improving Vision-Language Models (VLMs) on complex reasoning tasks. However, existing RL approaches often rely on human-annotated labels or task-specific heuristics to define verifiable rewards, both of which are costly and difficult to scale. We introduce VisPlay, a self-evolving RL framework that enables VLMs to autonomously improve their reasoning abilities using large amounts of unlabeled image data. Starting from a single base VLM, VisPlay assigns the model into two interacting roles: an Image-Conditioned Questioner that formulates challenging yet answerable visual questions, and a Multimodal Reasoner that generates silver responses. These roles are jointly trained with Group Relative Policy Optimization (GRPO), which incorporates diversity and difficulty rewards to balance the complexity of generated questions with the quality of the silver answers. VisPlay scales efficiently across two model families. When trained on Qwen2.5-VL and MiMo-VL, VisPlay achieves consistent improvements in visual reasoning, compositional generalization, and hallucination reduction across eight benchmarks, including MM-Vet and MMMU, demonstrating a scalable path toward self-evolving multimodal intelligence. The project page is available at https://bruno686.github.io/VisPlay/

VisPlay هو إطار تعلم معزز ذاتي التطور مصمم لتحسين نماذج الرؤية-اللغة (VLM) من خلال تمكينها من تحسين قدراتها على التفكير بشكل مستقل باستخدام كميات كبيرة من بيانات الصور غير المعلّمة. يعمل من خلال دورين: مُسْتَفْهَم مشروط بالصورة يقوم بصياغة أسئلة بصرية صعبة ولكن يمكن الإجابة عليها، ومفكر متعدد الوسائط يقوم بتوليد ردود فضية. يستخدم الإطار تحسين سياسة نسبية جماعية لتحقيق التوازن بين تعقيد الأسئلة وجودة الردود.

VisPlay es un marco de aprendizaje por refuerzo auto-evolutivo diseñado para mejorar los Modelos de Visión-Lenguaje (VLM) permitiendo que mejoren de forma autónoma sus capacidades de razonamiento utilizando grandes cantidades de datos de imágenes no etiquetadas. Funciona a través de dos roles: un Cuestionador Condicionado por Imágenes que genera preguntas visuales y un Razonador Multimodal que proporciona respuestas. El marco utiliza la Optimización de Políticas Relativas de Grupo para equilibrar la complejidad de las preguntas y la calidad de las respuestas.

VisPlay est un cadre d'apprentissage par renforcement auto-évolutif conçu pour améliorer les modèles de vision-langage (VLM) en leur permettant d'améliorer de manière autonome leurs capacités de raisonnement à l'aide de grandes quantités de données d'images non étiquetées. Il fonctionne à travers deux rôles : un questionneur conditionné par l'image qui génère des questions visuelles et un raisonneur multimodal qui fournit des réponses. Le cadre utilise l'optimisation de politique relative de groupe pour équilibrer la complexité des questions et la qualité des réponses.

VisPlay is a self-evolving reinforcement learning framework designed to enhance Vision-Language Models (VLMs) by enabling them to autonomously improve their reasoning capabilities using large amounts of unlabeled image data. It operates through two roles: an Image-Conditioned Questioner that generates visual questions and a Multimodal Reasoner that provides responses. The framework employs Group Relative Policy Optimization to balance question complexity and answer quality.

VisPlay: Self-Evolving Vision-Language Models from Images

arXiv:2511.13877v1 Announce Type: new 
Abstract: This paper proposes a new enhanced model architecture to perform classification of lumbar spine degeneration with DICOM images while using a hybrid approach, integrating EfficientNet and VGG19 together with custom-designed components. The proposed model is differentiated from traditional transfer learning methods as it incorporates a Pseudo-Newton Boosting layer along with a Sparsity-Induced Feature Reduction Layer that forms a multi-tiered framework, further improving feature selection and representation. The Pseudo-Newton Boosting layer makes smart variations of feature weights, with more detailed anatomical features, which are mostly left out in a transfer learning setup. In addition, the Sparsity-Induced Layer removes redundancy for learned features, producing lean yet robust representations for pathology in the lumbar spine. This architecture is novel as it overcomes the constraints in the traditional transfer learning approach, especially in the high-dimensional context of medical images, and achieves a significant performance boost, reaching a precision of 0.9, recall of 0.861, F1 score of 0.88, loss of 0.18, and an accuracy of 88.1%, compared to the baseline model, EfficientNet. This work will present the architectures, preprocessing pipeline, and experimental results. The results contribute to the development of automated diagnostic tools for medical images.

تقدم هذه الورقة نموذجًا معماريًا جديدًا لتصنيف تدهور العمود الفقري القطني باستخدام صور DICOM. يجمع هذا النهج الهجين بين EfficientNet و VGG19 مع مكونات مصممة خصيصًا، مما يميزه عن طرق التعلم بالنقل التقليدية. يتضمن طبقة تعزيز بوساطة نيوتن الزائفة التي تعزز من تنويعات أوزان الميزات وطبقة تقليل الميزات المستندة إلى الندرة التي تقلل من التكرار في الميزات المتعلمة. تهدف هذه البنية الجديدة إلى تحسين تمثيل الميزات التشريحية التي غالبًا ما يتم تجاهلها في الإعدادات التقليدية، مما يؤدي إلى اكتشاف أكثر دقة للمرض.

El artículo presenta una nueva arquitectura de modelo para clasificar la degeneración de la columna lumbar utilizando imágenes DICOM. Este enfoque híbrido combina EfficientNet y VGG19 con componentes personalizados, diferenciándose de los métodos tradicionales de aprendizaje por transferencia. Incluye una capa de refuerzo Pseudo-Newton que mejora las variaciones de los pesos de las características y una capa de reducción de características inducida por la escasez que minimiza la redundancia en las características aprendidas. Esta arquitectura innovadora busca mejorar la representación de carac…

Cet article propose une nouvelle architecture de modèle pour classer la dégénérescence de la colonne lombaire à l'aide d'images DICOM. Cette approche hybride combine EfficientNet et VGG19 avec des composants personnalisés, la distinguant des méthodes traditionnelles d'apprentissage par transfert. Elle comprend une couche de renforcement Pseudo-Newton qui améliore les variations des poids des caractéristiques et une couche de réduction de caractéristiques induite par la parcimonie qui minimise la redondance des caractéristiques apprises. Cette architecture innovante vise à améliorer la représen…

The paper presents a novel model architecture for classifying lumbar spine degeneration using DICOM images. This hybrid approach combines EfficientNet and VGG19 with custom components, distinguishing it from traditional transfer learning methods. It features a Pseudo-Newton Boosting layer that enhances feature weight variations and a Sparsity-Induced Feature Reduction Layer that minimizes redundancy in learned features. This innovative architecture aims to improve the representation of anatomical features often overlooked in conventional setups, leading to more accurate pathology detection.

Hybrid Convolution Neural Network Integrated with Pseudo-Newton Boosting for Lumbar Spine Degeneration Detection

arXiv:2511.15406v1 Announce Type: cross 
Abstract: Reliable semantic segmentation is essential for clinical decision making, yet deep models rarely provide explicit statistical guarantees on their errors. We introduce a simple post-hoc framework that constructs confidence masks with distribution-free, image-level control of false-positive predictions. Given any pretrained segmentation model, we define a nested family of shrunken masks obtained either by increasing the score threshold or by applying morphological erosion. A labeled calibration set is used to select a single shrink parameter via conformal prediction, ensuring that, for new images that are exchangeable with the calibration data, the proportion of false positives retained in the confidence mask stays below a user-specified tolerance with high probability. The method is model-agnostic, requires no retraining, and provides finite-sample guarantees regardless of the underlying predictor. Experiments on a polyp-segmentation benchmark demonstrate target-level empirical validity. Our framework enables practical, risk-aware segmentation in settings where over-segmentation can have clinical consequences. Code at https://github.com/deel-ai-papers/conseco.

تم تقديم إطار عمل جديد للتحكم في الإيجابيات الكاذبة في تقسيم الصور، مما يعزز موثوقية التقسيم الدلالي في اتخاذ القرارات السريرية. تستخدم هذه الطريقة المستقلة عن النموذج التنبؤ المتوافق لإنشاء أقنعة ثقة تحافظ على مستوى محدد مسبقًا من الإيجابيات الكاذبة، دون الحاجة إلى إعادة التدريب. تُظهر الطريقة ضمانات عالية الاحتمالية للصور الجديدة، مما يمثل تقدمًا كبيرًا في التصوير الطبي.

Se ha introducido un nuevo marco para controlar los falsos positivos en la segmentación de imágenes, mejorando la fiabilidad de la segmentación semántica en la toma de decisiones clínicas. Este enfoque independiente del modelo utiliza la predicción conformal para crear máscaras de confianza que mantienen una tolerancia definida por el usuario para los falsos positivos, sin necesidad de reentrenamiento. El método demuestra garantías de alta probabilidad para nuevas imágenes, representando un avance significativo en la imagen médica.

Un nouveau cadre pour contrôler les faux positifs dans la segmentation d'images a été introduit, améliorant la fiabilité de la segmentation sémantique dans la prise de décision clinique. Cette approche indépendante du modèle utilise la prédiction conforme pour créer des masques de confiance qui maintiennent une tolérance définie par l'utilisateur pour les faux positifs, sans nécessiter de réentraînement. La méthode démontre des garanties de haute probabilité pour de nouvelles images, représentant une avancée significative dans l'imagerie médicale.

A new framework for controlling false positives in image segmentation has been introduced, enhancing the reliability of semantic segmentation in clinical decision-making. This model-agnostic approach utilizes conformal prediction to create confidence masks that maintain a user-defined tolerance for false positives, without requiring retraining. The method demonstrates high probability guarantees for new images, making it a significant advancement in medical imaging.

Controlling False Positives in Image Segmentation via Conformal Prediction

arXiv:2504.07503v2 Announce Type: replace 
Abstract: Event cameras asynchronously capture brightness changes with microsecond latency, offering exceptional temporal precision but suffering from severe noise and signal inconsistencies. Unlike conventional signals, events carry state information through polarities and process information through inter-event time intervals. However, existing event filters often ignore the latter, producing outputs that are sparser than the raw input and limiting the reconstruction of continuous irradiance dynamics. We propose the Event Density Flow Filter (EDFilter), a framework that models event generation as threshold-crossing probability fluxes arising from the stochastic diffusion of irradiance trajectories. EDFilter performs nonparametric, kernel-based estimation of probability flux and reconstructs the continuous event density flow using an O(1) recursive solver, enabling real-time processing. The Rotary Event Dataset (RED), featuring microsecond-resolution ground-truth irradiance flow under controlled illumination is also presented for event quality evaluation. Experiments demonstrate that EDFilter achieves high-fidelity, physically interpretable event denoising and motion reconstruction.

تقدم الورقة فلاتر تدفق كثافة الأحداث (EDFilter)، وهو إطار مبتكر لتصفية تدفقات الأحداث يستخدم تقدير تدفق الاحتمالات لتحسين معالجة بيانات كاميرات الأحداث. تتناول هذه الطريقة قيود الفلاتر الحالية من خلال نمذجة توليد الأحداث كتيارات احتمالية تتجاوز العتبات، مما يسمح بإعادة بناء تدفق كثافة الأحداث المستمر في الوقت الحقيقي. يهدف الإطار إلى تحسين دقة وكفاءة أنظمة الرؤية المعتمدة على الأحداث.

El artículo presenta el Filtro de Flujo de Densidad de Eventos (EDFilter), un marco novedoso para el filtrado de flujos de eventos que utiliza la estimación de flujo de probabilidad para mejorar el procesamiento de datos de cámaras de eventos. Este enfoque aborda las limitaciones de los filtros de eventos existentes al modelar la generación de eventos como flujos de probabilidad que cruzan umbrales, permitiendo la reconstrucción en tiempo real del flujo de densidad de eventos continuo. El marco busca mejorar la precisión y eficiencia de los sistemas de visión basados en eventos.

L'article présente le filtre de flux de densité d'événements (EDFilter), un cadre novateur pour le filtrage de flux d'événements qui utilise l'estimation de flux de probabilité pour améliorer le traitement des données des caméras d'événements. Cette approche répond aux limitations des filtres d'événements existants en modélisant la génération d'événements comme des flux de probabilité franchissant des seuils, permettant ainsi la reconstruction en temps réel du flux de densité d'événements continu. Le cadre vise à améliorer la précision et l'efficacité des systèmes de vision basés sur des événe…

The paper introduces the Event Density Flow Filter (EDFilter), a novel framework for event stream filtering that utilizes probability flux estimation to enhance the processing of event camera data. This approach addresses the limitations of existing event filters by modeling event generation as threshold-crossing probability fluxes, allowing for real-time reconstruction of continuous event density flow. The framework aims to improve the accuracy and efficiency of event-based vision systems.

Event Stream Filtering via Probability Flux Estimation

arXiv:2511.15499v1 Announce Type: new 
Abstract: Autoregressive models have recently shown great promise in visual generation by leveraging discrete token sequences akin to language modeling. However, existing approaches often suffer from inefficiency, either due to token-by-token decoding or the complexity of multi-scale representations. In this work, we introduce Expanding Autoregressive Representation (EAR), a novel generation paradigm that emulates the human visual system's center-outward perception pattern. EAR unfolds image tokens in a spiral order from the center and progressively expands outward, preserving spatial continuity and enabling efficient parallel decoding. To further enhance flexibility and speed, we propose a length-adaptive decoding strategy that dynamically adjusts the number of tokens predicted at each step. This biologically inspired design not only reduces computational cost but also improves generation quality by aligning the generation order with perceptual relevance. Extensive experiments on ImageNet demonstrate that EAR achieves state-of-the-art trade-offs between fidelity and efficiency on single-scale autoregressive models, setting a new direction for scalable and cognitively aligned autoregressive image generation.

تقدم الورقة مفهوم تمثيل التوليد التلقائي الموسع (EAR)، وهو نموذج جديد لتوليد الصور يحاكي نظام الرؤية البشري من خلال إدراكه من المركز إلى الخارج. تعمل هذه الطريقة على تحسين الكفاءة من خلال نشر رموز الصورة بترتيب حلزوني، مما يسمح بفك تشفير متوازي ويحافظ على الاستمرارية المكانية. بالإضافة إلى ذلك، يتم اقتراح استراتيجية فك تشفير قابلة للتكيف مع الطول لتعزيز المرونة والسرعة، مما يقلل من التكاليف الحسابية ويحسن جودة التوليد.

El artículo presenta la Representación Autoregresiva Expandida (EAR), un nuevo paradigma para la generación visual que imita el sistema visual humano al percibir de manera centrípeta. Este método mejora la eficiencia al desplegar tokens de imagen en un orden espiral, permitiendo la decodificación paralela y preservando la continuidad espacial. Además, se propone una estrategia de decodificación adaptativa en longitud para mejorar la flexibilidad y velocidad, reduciendo así los costos computacionales y mejorando la calidad de generación.

Cet article présente la Représentation Autoregressive Élargie (EAR), un nouveau paradigme pour la génération visuelle qui imite le système visuel humain en percevant de manière centripète. Cette méthode améliore l'efficacité en déroulant les tokens d'image dans un ordre spiralé, permettant un décodage parallèle et préservant la continuité spatiale. De plus, une stratégie de décodage adaptative en longueur est proposée pour améliorer la flexibilité et la rapidité, réduisant ainsi les coûts computationnels et améliorant la qualité de génération.

The paper introduces Expanding Autoregressive Representation (EAR), a new paradigm for visual generation that mimics the human visual system's center-outward perception. This method improves efficiency by unfolding image tokens in a spiral order, allowing for parallel decoding and preserving spatial continuity. Additionally, a length-adaptive decoding strategy is proposed to enhance flexibility and speed, ultimately reducing computational costs and improving generation quality.

Governance-Ready Small Language Models for Medical Imaging: Prompting, Abstention, and PACS Integration

Was this article worth reading? Share it