arXiv:2511.19735v1 Announce Type: cross 
Abstract: Randomized controlled trials (RCTs) have been the cornerstone of clinical evidence; however, their cost, duration, and restrictive eligibility criteria limit power and external validity. Studies using real-world data (RWD), historically considered less reliable for establishing causality, are now recognized to be important for generating real-world evidence (RWE). In parallel, artificial intelligence and machine learning (AI/ML) are being increasingly used throughout the drug development process, providing scalability and flexibility but also presenting challenges in interpretability and rigor that traditional statistics do not face. This Perspective argues that the future of evidence generation will not depend on RCTs versus RWD, or statistics versus AI/ML, but on their principled integration. To this end, a causal roadmap is needed to clarify inferential goals, make assumptions explicit, and ensure transparency about tradeoffs. We highlight key objectives of integrative evidence synthesis, including transporting RCT results to broader populations, embedding AI-assisted analyses within RCTs, designing hybrid controlled trials, and extending short-term RCTs with long-term RWD. We also outline future directions in privacy-preserving analytics, uncertainty quantification, and small-sample methods. By uniting statistical rigor with AI/ML innovation, integrative approaches can produce robust, transparent, and policy-relevant evidence, making them a key component of modern regulatory science.

تركز وجهة نظر جديدة حول توليد الأدلة على دمج التجارب السريرية العشوائية (RCTs) والبيانات الواقعية (RWD) والذكاء الاصطناعي وتعلم الآلة (AI/ML) والإحصائيات التقليدية. تتناول هذه المقاربة قيود التجارب السريرية العشوائية، مثل التكاليف العالية ومعايير الأهلية الصارمة، مع الاعتراف بأهمية البيانات الواقعية المتزايدة في توليد الأدلة الواقعية (RWE).

Una nueva perspectiva sobre la generación de evidencia enfatiza la integración de ensayos controlados aleatorios (ECA), datos del mundo real (DWR), inteligencia artificial y aprendizaje automático (IA/AA), y estadísticas tradicionales. Este enfoque aborda las limitaciones de los ECA, como los altos costos y los criterios de elegibilidad restrictivos, al mismo tiempo que reconoce la creciente importancia de los DWR en la generación de evidencia del mundo real (EWR).

Une nouvelle perspective sur la génération de preuves met l'accent sur l'intégration des essais contrôlés randomisés (ECR), des données du monde réel (DWR), de l'intelligence artificielle et de l'apprentissage automatique (IA/AA), ainsi que des statistiques traditionnelles. Cette approche répond aux limites des ECR, telles que les coûts élevés et les critères d'éligibilité restrictifs, tout en reconnaissant l'importance croissante des DWR dans la génération de preuves du monde réel (PWR).

A new perspective on evidence generation emphasizes the integration of randomized controlled trials (RCTs), real-world data (RWD), artificial intelligence and machine learning (AI/ML), and traditional statistics. This approach addresses the limitations of RCTs, such as high costs and restrictive eligibility, while recognizing the growing importance of RWD in generating real-world evidence (RWE).

Integrating RCTs, RWD, AI/ML and Statistics: Next-Generation Evidence Synthesis

arXiv:2511.20489v1 Announce Type: cross 
Abstract: Real-world AI/ML workflows often apply inference computations to feature vectors joined from multiple datasets. To avoid the redundant AI/ML computations caused by repeated data records in the join's output, factorized ML has been proposed to decompose ML computations into sub-computations to be executed on each normalized dataset. However, there is insufficient discussion on how factorized ML could impact AI/ML inference over multi-way joins. To address the limitations, we propose a novel declarative InferF system, focusing on the factorization of arbitrary inference workflows represented as analyzable expressions over the multi-way joins. We formalize our problem to flexibly push down partial factorized computations to qualified nodes in the join tree to minimize the overall inference computation and join costs and propose two algorithms to resolve the problem: (1) a greedy algorithm based on a per-node cost function that estimates the influence on overall latency if a subset of factorized computations is pushed to a node, and (2) a genetic algorithm for iteratively enumerating and evaluating promising factorization plans. We implement InferF on Velox, an open-sourced database engine from Meta, evaluate it on real-world datasets, observed up to 11.3x speedups, and systematically summarized the factors that determine when factorized ML can benefit AI/ML inference workflows.

تم اقتراح نظام جديد يسمى InferF لتحسين سير العمل في الذكاء الاصطناعي/تعلم الآلة من خلال تحليل حسابات الاستدلال على الانضمامات متعددة مجموعات البيانات. تهدف هذه الطريقة إلى تقليل الحسابات المكررة من خلال تفكيك مهام التعلم الآلي إلى حسابات فرعية يتم تنفيذها على مجموعات البيانات العادية، مما يعالج القيود المفروضة على الأساليب الحالية.

Se ha propuesto un nuevo sistema llamado InferF para mejorar los flujos de trabajo de IA/ML mediante la factorización de cálculos de inferencia sobre uniones de múltiples conjuntos de datos. Este enfoque busca minimizar los cálculos redundantes al descomponer las tareas de aprendizaje automático en subcálculos ejecutados en conjuntos de datos normalizados, abordando así las limitaciones de los métodos existentes.

Un nouveau système appelé InferF a été proposé pour améliorer les flux de travail en IA/ML en factorisant les calculs d'inférence sur des jointures multi-ensembles de données. Cette approche vise à minimiser les calculs redondants en décomposant les tâches d'apprentissage automatique en sous-calculs exécutés sur des ensembles de données normalisés, abordant ainsi les limites des méthodes existantes.

A novel system called InferF has been proposed to enhance AI/ML workflows by factorizing inference computations over multi-way joins of datasets. This approach aims to minimize redundant computations by decomposing machine learning tasks into sub-computations executed on normalized datasets, addressing the limitations of existing methods.

InferF: Declarative Factorization of AI/ML Inferences over Joins

arXiv:2511.17505v1 Announce Type: cross 
Abstract: To keep modern Radio Access Networks (RAN) running smoothly, operators need to spot the real-world triggers behind Service-Level Agreement (SLA) breaches well before customers feel them. We introduce an AI/ML pipeline that does two things most tools miss: (1) finds the likely root-cause indicators and (2) reveals the exact order in which those events unfold. We start by labeling network data: records linked to past SLA breaches are marked `abnormal', and everything else `normal'. Our model then learns the causal chain that turns normal behavior into a fault. In Monte Carlo tests the approach pinpoints the correct trigger sequence with high precision and scales to millions of data points without loss of speed. These results show that high-resolution, causally ordered insights can move fault management from reactive troubleshooting to proactive prevention.

تم تقديم خط أنابيب جديد يعتمد على الذكاء الاصطناعي/تعلم الآلة لتحسين تتبع الأعطال في شبكات الوصول اللاسلكي (RAN) من خلال تحديد المحفزات الحقيقية وراء انتهاكات اتفاقيات مستوى الخدمة (SLA) قبل أن تؤثر على العملاء. يقوم النموذج بتصنيف بيانات الشبكة على أنها 'غير طبيعية' أو 'طبيعية' ويتعلم السلسلة السببية التي تؤدي إلى الأعطال، محققًا دقة عالية في تحديد تسلسلات المحفزات خلال اختبارات مونت كارلو.

Se ha introducido un nuevo pipeline de IA/ML para mejorar el seguimiento de fallos en las Redes de Acceso Radio (RAN) al identificar los desencadenantes reales detrás de las violaciones de los Acuerdos de Nivel de Servicio (SLA) antes de que impacten a los clientes. El modelo etiqueta los datos de la red como 'anormales' o 'normales' y aprende la cadena causal que lleva a los fallos, logrando una alta precisión en la identificación de secuencias de desencadenamiento durante las pruebas de Monte Carlo.

Un nouveau pipeline d'IA/ML a été introduit pour améliorer le suivi des défauts dans les réseaux d'accès radio (RAN) en identifiant les déclencheurs réels des violations des accords de niveau de service (SLA) avant qu'elles n'impactent les clients. Le modèle étiquette les données réseau comme 'anormales' ou 'normales' et apprend la chaîne causale menant aux défauts, atteignant une haute précision dans l'identification des séquences de déclenchement lors de tests de Monte Carlo.

A new AI/ML pipeline has been introduced to enhance fault tracking in Radio Access Networks (RAN) by identifying real-world triggers behind Service-Level Agreement (SLA) breaches before they impact customers. The model labels network data as 'abnormal' or 'normal' and learns the causal chain leading to faults, achieving high precision in pinpointing trigger sequences during Monte Carlo tests.

Causal Intervention Sequence Analysis for Fault Tracking in Radio Access Networks

arXiv:2402.07388v4 Announce Type: replace-cross 
Abstract: Algorithm evaluation and comparison are fundamental questions in machine learning and statistics -- how well does an algorithm perform at a given modeling task, and which algorithm performs best? Many methods have been developed to assess algorithm performance, often based around cross-validation type strategies, retraining the algorithm of interest on different subsets of the data and assessing its performance on the held-out data points. Despite the broad use of such procedures, the theoretical properties of these methods are not yet fully understood. In this work, we explore some fundamental limits for answering these questions with limited amounts of data. In particular, we make a distinction between two questions: how good is an algorithm A at the problem of learning from a training set of size n, versus, how good is a particular fitted model produced by running A on a particular training data set of size $n$? Our main results prove that, for any test that treats the algorithm A as a ``black box'' (i.e., we can only study the behavior of A empirically), there is a fundamental limit on our ability to carry out inference on the performance of A, unless the number of available data points $N$ is many times larger than the evaluation sample size $n$ of interest. On the other hand, evaluating the performance of a particular fitted model can be easy when evaluating an algorithm is hard. We also ask whether an assumption of algorithmic stability might be sufficient to circumvent this hardness result. Surprisingly, we find that the same hardness result still holds for the problem of evaluating the performance of A, aside from a high-stability regime where fitted models are essentially nonrandom. Finally, we also establish similar hardness results for the problem of comparing multiple algorithms.

دراسة حديثة نُشرت على arXiv تستكشف حدود الاختبارات الخالية من الافتراضات لتقييم أداء الخوارزميات في التعلم الآلي والإحصاء. تميز الدراسة بين تقييم قدرة الخوارزمية على التعلم من مجموعة تدريب معينة وتقييم نموذج محدد تم إنتاجه بواسطة تلك الخوارزمية. تسلط الأبحاث الضوء على الفجوات النظرية في فهم هذه الأساليب التقييمية، خاصة عندما تكون البيانات محدودة.

Un estudio reciente publicado en arXiv explora las limitaciones de las pruebas sin suposiciones para evaluar el rendimiento de algoritmos en aprendizaje automático y estadísticas. Se distingue entre evaluar la capacidad de un algoritmo para aprender de un conjunto de entrenamiento y evaluar un modelo específico producido por ese algoritmo. La investigación destaca las lagunas teóricas en la comprensión de estos métodos de evaluación, especialmente cuando los datos son limitados.

Une étude récente publiée sur arXiv explore les limites des tests sans hypothèse pour évaluer la performance des algorithmes en apprentissage automatique et en statistiques. Elle fait une distinction entre l'évaluation de la capacité d'un algorithme à apprendre à partir d'un ensemble d'entraînement et l'évaluation d'un modèle spécifique produit par cet algorithme. La recherche met en lumière les lacunes théoriques dans la compréhension de ces méthodes d'évaluation, en particulier lorsque les données sont limitées.

A recent study published on arXiv explores the limitations of assumption-free tests for evaluating algorithm performance in machine learning and statistics. It distinguishes between assessing an algorithm's ability to learn from a training set and evaluating a specific model produced by that algorithm. The research highlights the theoretical gaps in understanding these evaluation methods, particularly when data is limited.

The Limits of Assumption-free Tests for Algorithm Performance

arXiv:2209.07396v3 Announce Type: replace 
Abstract: Score-based divergences have been widely used in machine learning and statistics applications. Despite their empirical success, a blindness problem has been observed when using these for multi-modal distributions. In this work, we discuss the blindness problem and propose a new family of divergences that can mitigate the blindness problem. We illustrate our proposed divergence in the context of density estimation and report improved performance compared to traditional approaches.

أظهرت دراسة حديثة وجود مشكلة عمياء في التباينات المعتمدة على الدرجات المستخدمة في التعلم الآلي والإحصاءات، مما يؤثر بشكل خاص على التوزيعات متعددة الأوضاع. تقترح الدراسة عائلة جديدة من التباينات تهدف إلى التخفيف من هذه المشكلة، مع عرض أداء محسّن في مهام تقدير الكثافة مقارنة بالطرق التقليدية.

Un estudio reciente ha identificado un problema de ceguera en las divergencias basadas en puntuaciones utilizadas en aprendizaje automático y estadísticas, afectando particularmente a las distribuciones multimodales. La investigación propone una nueva familia de divergencias destinadas a mitigar este problema, demostrando un rendimiento mejorado en tareas de estimación de densidad en comparación con los métodos tradicionales.

Une étude récente a identifié un problème de cécité dans les divergences basées sur les scores utilisées en apprentissage automatique et en statistiques, affectant particulièrement les distributions multimodales. La recherche propose une nouvelle famille de divergences visant à atténuer ce problème, démontrant une performance améliorée dans les tâches d'estimation de densité par rapport aux méthodes traditionnelles.

A recent study has identified a blindness problem in score-based divergences used in machine learning and statistics, particularly affecting multi-modal distributions. The research proposes a new family of divergences aimed at mitigating this issue, demonstrating improved performance in density estimation tasks compared to traditional methods.

Integrating RCTs, RWD, AI/ML and Statistics: Next-Generation Evidence Synthesis

Was this article worth reading? Share it

Sourcely

Media Workbench AI

Redreach AI