arXiv:2511.15487v1 Announce Type: new 
Abstract: Implicit Neural Representations (INRs) parameterize continuous signals via multilayer perceptrons (MLPs), enabling compact, resolution-independent modeling for tasks like image, audio, and 3D reconstruction. However, fitting high-resolution signals demands optimizing over millions of coordinates, incurring prohibitive computational costs. To address it, we propose NTK-Guided Implicit Neural Teaching (NINT), which accelerates training by dynamically selecting coordinates that maximize global functional updates. Leveraging the Neural Tangent Kernel (NTK), NINT scores examples by the norm of their NTK-augmented loss gradients, capturing both fitting errors and heterogeneous leverage (self-influence and cross-coordinate coupling). This dual consideration enables faster convergence compared to existing methods. Through extensive experiments, we demonstrate that NINT significantly reduces training time by nearly half while maintaining or improving representation quality, establishing state-of-the-art acceleration among recent sampling-based strategies.

تعتبر طريقة التعليم العصبي الضمني الموجه بواسطة NTK (NINT) مقترحًا لتحسين تدريب التمثيلات العصبية الضمنية (INRs) من خلال تحسين اختيار الإحداثيات لتعزيز الكفاءة الحاسوبية. من خلال استخدام النواة المماسية العصبية (NTK)، تقوم NINT بتقييم الأمثلة بناءً على تدرجات خسائرها، مما يسمح بتقارب أسرع في التدريب. تشير النتائج التجريبية إلى أن NINT يمكن أن تقلل من وقت التدريب بمقدار النصف تقريبًا مع الحفاظ على الأداء.

La enseñanza neural implícita guiada por NTK (NINT) es un método propuesto que mejora el entrenamiento de las Representaciones Neurales Implícitas (INRs) al optimizar la selección de coordenadas para mejorar la eficiencia computacional. Al utilizar el Núcleo Tangente Neural (NTK), NINT puntúa ejemplos según sus gradientes de pérdida, lo que permite una convergencia más rápida en el entrenamiento. Los resultados experimentales indican que NINT puede reducir el tiempo de entrenamiento casi a la mitad mientras mantiene el rendimiento.

L'enseignement neural implicite guidé par NTK (NINT) est une méthode proposée qui améliore l'entraînement des représentations neuronales implicites (INRs) en optimisant la sélection des coordonnées pour améliorer l'efficacité computationnelle. En utilisant le noyau tangent neural (NTK), NINT évalue les exemples en fonction de leurs gradients de perte, permettant une convergence plus rapide lors de l'entraînement. Les résultats expérimentaux montrent que NINT peut réduire le temps d'entraînement de près de moitié tout en maintenant les performances.

NTK-Guided Implicit Neural Teaching (NINT) is a proposed method that enhances the training of Implicit Neural Representations (INRs) by optimizing coordinate selection to improve computational efficiency. By utilizing the Neural Tangent Kernel (NTK), NINT scores examples based on their loss gradients, allowing for faster convergence in training. Experimental results indicate that NINT can reduce training time by nearly half while maintaining performance.

NTK-Guided Implicit Neural Teaching

arXiv:2601.06224v2 Announce Type: replace 
Abstract: While Multimodal Large Language Models (MLLMs) have achieved remarkable success across diverse tasks, their practical deployment is severely hindered by hallucination issues, which become particularly acute during Reinforcement Learning (RL) optimization. This paper systematically analyzes the root causes of hallucinations in MLLMs under RL training, identifying three critical factors: (1) an over-reliance on chained visual reasoning, where inaccurate initial descriptions or redundant information anchor subsequent inferences to incorrect premises; (2) insufficient exploration diversity during policy optimization, leading the model to generate overly confident but erroneous outputs; and (3) destructive conflicts between training samples, where Neural Tangent Kernel (NTK) similarity causes false associations and unstable parameter updates. To address these challenges, we propose a comprehensive framework comprising three core modules. First, we enhance visual localization by introducing dedicated planning and captioning stages before the reasoning phase, employing a quality-based caption reward to ensure accurate initial anchoring. Second, to improve exploration, we categorize samples based on the mean and variance of their reward distributions, prioritizing samples with high variance to focus the model on diverse and informative data. Finally, to mitigate sample interference, we regulate NTK similarity by grouping sample pairs and applying an InfoNCE loss to push overly similar pairs apart and pull dissimilar ones closer, thereby guiding gradient interactions toward a balanced range. Experimental results demonstrate that our proposed method significantly reduces hallucination rates and effectively enhances the inference accuracy of MLLMs.

أظهرت دراسة حديثة إطارًا يهدف إلى التخفيف من مشاكل الهلوسة في نماذج اللغة متعددة الوسائط (MLLM) أثناء تحسين التعلم المعزز (RL). تحدد الدراسة عوامل رئيسية تسهم في الهلوسة، بما في ذلك الاعتماد المفرط على التفكير البصري ونقص تنوع الاستكشاف. يتضمن الإطار المقترح وحدات للتغذية الراجعة على التسميات، وأخذ العينات الواعية للتنوع، وتنظيم النزاعات لتحسين موثوقية النموذج.

Un estudio reciente ha presentado un marco destinado a mitigar los problemas de alucinaciones en los Modelos de Lenguaje Multimodal (MLLM) durante la optimización del Aprendizaje por Refuerzo (RL). La investigación identifica factores clave que contribuyen a las alucinaciones, incluyendo la dependencia excesiva del razonamiento visual y la insuficiente diversidad de exploración. El marco propuesto incorpora módulos para la retroalimentación de leyendas, muestreo consciente de la diversidad y regularización de conflictos para mejorar la fiabilidad del modelo.

Une étude récente a introduit un cadre visant à atténuer les problèmes d'hallucination dans les modèles de langage multimodal (MLLM) lors de l'optimisation par apprentissage par renforcement (RL). La recherche identifie des facteurs clés contribuant aux hallucinations, notamment la dépendance excessive au raisonnement visuel et l'insuffisance de diversité d'exploration. Le cadre proposé intègre des modules pour le retour d'information sur les légendes, l'échantillonnage conscient de la diversité et la régularisation des conflits pour améliorer la fiabilité du modèle.

A recent study has introduced a framework aimed at mitigating hallucination issues in Multimodal Large Language Models (MLLMs) during Reinforcement Learning (RL) optimization. The research identifies key factors contributing to hallucinations, including over-reliance on visual reasoning and insufficient exploration diversity. The proposed framework incorporates modules for caption feedback, diversity-aware sampling, and conflict regularization to enhance model reliability.

Ground What You See: Hallucination-Resistant MLLMs via Caption Feedback, Diversity-Aware Sampling, and Conflict Regularization

One More Thing in AI – Your Shortcut to AI Mastery

NTK-Guided Implicit Neural Teaching

Was this article worth reading? Share it

One More Thing in AI

LucidQuery AI

The Visualizer

Hypertune

IntelliQ

Blunge

Ready to build your own newsroom?