arXiv:2511.01352v1 Announce Type: new 
Abstract: In this paper, we present a new algorithm, MiniFool, that implements physics-inspired adversarial attacks for testing neural network-based classification tasks in particle and astroparticle physics. While we initially developed the algorithm for the search for astrophysical tau neutrinos with the IceCube Neutrino Observatory, we apply it to further data from other science domains, thus demonstrating its general applicability. Here, we apply the algorithm to the well-known MNIST data set and furthermore, to Open Data data from the CMS experiment at the Large Hadron Collider. The algorithm is based on minimizing a cost function that combines a $\chi^2$ based test-statistic with the deviation from the desired target score. The test statistic quantifies the probability of the perturbations applied to the data based on the experimental uncertainties. For our studied use cases, we find that the likelihood of a flipped classification differs for both the initially correctly and incorrectly classified events. When testing changes of the classifications as a function of an attack parameter that scales the experimental uncertainties, the robustness of the network decision can be quantified. Furthermore, this allows testing the robustness of the classification of unlabeled experimental data.

تقديم MiniFool يمثل تقدمًا كبيرًا في مجال الهجمات العدائية على الشبكات العصبية العميقة، خاصة في سياق فيزياء الجسيمات والفيزياء الفلكية. لا يعزز هذا الخوارزم الجديد فقط البحث عن النيوترينوات التاو الفلكية في مرصد النيوترينوات IceCube، بل يظهر أيضًا تنوعه من خلال تطبيقه في مجالات علمية أخرى. هذا التطور مهم لأنه يفتح آفاقًا جديدة لاختبار وتحسين نماذج الشبكات العصبية، مما يساهم في تحقيق أبحاث علمية أكثر قوة وموثوقية.

La introducción de MiniFool marca un avance significativo en el ámbito de los ataques adversariales en redes neuronales profundas, especialmente en el contexto de la física de partículas y astrofísica. Este innovador algoritmo no solo mejora la búsqueda de neutrinos tau astrofísicos en el Observatorio de Neutrinos IceCube, sino que también demuestra su versatilidad al ser aplicable en varios dominios científicos. Este desarrollo es crucial, ya que abre nuevas avenidas para probar y mejorar los modelos de redes neuronales, contribuyendo así a una investigación científica más robusta y confiable.

L'introduction de MiniFool représente une avancée significative dans le domaine des attaques adversariales sur les réseaux de neurones profonds, en particulier dans le contexte de la physique des particules et de l'astrophysique. Cet algorithme innovant améliore non seulement la recherche des neutrinos tau astrophysiques à l'Observatoire de neutrinos IceCube, mais démontre également sa polyvalence en étant applicable à divers domaines scientifiques. Ce développement est crucial car il ouvre de nouvelles voies pour tester et améliorer les modèles de réseaux de neurones, contribuant ainsi à une recherche scientifique plus robuste et fiable.

The introduction of MiniFool marks a significant advancement in the realm of adversarial attacks on deep neural networks, particularly in the context of particle and astroparticle physics. This innovative algorithm not only enhances the search for astrophysical tau neutrinos at the IceCube Neutrino Observatory but also showcases its versatility by being applicable across various scientific domains. This development is crucial as it opens new avenues for testing and improving neural network models, ultimately contributing to more robust and reliable scientific research.

MiniFool - Physics-Constraint-Aware Minimizer-Based Adversarial Attacks in Deep Neural Networks

arXiv:2511.18567v1 Announce Type: new 
Abstract: The Forward-Forward (FF) algorithm offers a biologically plausible alternative to backpropagation, enabling neural networks to learn through local updates. However, FF's efficacy relies heavily on the definition of "goodness", which is a scalar measure of neural activity. While current implementations predominantly utilize a simple sum-of-squares metric, it remains unclear if this default choice is optimal. To address this, we benchmarked 21 distinct goodness functions across four standard image datasets (MNIST, FashionMNIST, CIFAR-10, STL-10), evaluating classification accuracy, energy consumption, and carbon footprint. We found that certain alternative goodness functions inspired from various domains significantly outperform the standard baseline. Specifically, \texttt{game\_theoretic\_local} achieved 97.15\% accuracy on MNIST, \texttt{softmax\_energy\_margin\_local} reached 82.84\% on FashionMNIST, and \texttt{triplet\_margin\_local} attained 37.69\% on STL-10. Furthermore, we observed substantial variability in computational efficiency, highlighting a critical trade-off between predictive performance and environmental cost. These findings demonstrate that the goodness function is a pivotal hyperparameter in FF design. We release our code on \href{https://github.com/aryashah2k/In-Search-of-Goodness}{Github} for reference and reproducibility.

يقدم خوارزمية Forward-Forward (FF) بديلاً بيولوجيًا معقولًا للتراجع التقليدي في الشبكات العصبية، حيث يركز على التحديثات المحلية من خلال مقياس عددي لـ 'الخير'. كشفت الاختبارات الأخيرة لـ 21 وظيفة خير مختلفة عبر أربعة مجموعات بيانات صور قياسية أن بعض البدائل تتفوق بشكل كبير على المقياس التقليدي لمجموع المربعات، مع تحسينات ملحوظة في الدقة على مجموعات بيانات مثل MNIST وFashionMNIST.

El algoritmo Forward-Forward (FF) ofrece una alternativa biológicamente plausible a la retropropagación tradicional en redes neuronales, enfocándose en actualizaciones locales a través de una medida escalar de 'bondad'. Recientes pruebas de 21 funciones de bondad distintas en cuatro conjuntos de datos de imágenes estándar revelaron que ciertas alternativas superan significativamente la métrica convencional de suma de cuadrados, con mejoras notables en precisión en conjuntos de datos como MNIST y FashionMNIST.

L'algorithme Forward-Forward (FF) propose une alternative biologiquement plausible à la rétropropagation traditionnelle dans les réseaux de neurones, en se concentrant sur des mises à jour locales via une mesure scalaire de la 'bonté'. Des tests récents de 21 fonctions de bonté distinctes sur quatre ensembles de données d'images standard ont révélé que certaines alternatives surpassent de manière significative la métrique conventionnelle de somme des carrés, avec des améliorations notables de précision sur des ensembles de données comme MNIST et FashionMNIST.

The Forward-Forward (FF) algorithm presents a biologically plausible alternative to traditional backpropagation in neural networks, focusing on local updates through a scalar measure of 'goodness'. Recent benchmarking of 21 distinct goodness functions across four standard image datasets revealed that certain alternatives significantly outperform the conventional sum-of-squares metric, with notable accuracy improvements on datasets like MNIST and FashionMNIST.

In Search of Goodness: Large Scale Benchmarking of Goodness Functions for the Forward-Forward Algorithm

arXiv:2511.17638v1 Announce Type: new 
Abstract: Modern artificial intelligence systems depend heavily on large datasets for both training and transferring knowledge between models. Knowledge distillation, transfer learning, and dataset distillation have made such transfers more efficient, yet they remain fundamentally data-driven: a teacher must produce examples, logits, or gradients for a student to learn. In this work, we introduce Model-to-Model Knowledge Transmission (M2KT), a novel paradigm for data-free conceptual transfer between neural networks. M2KT enables models to exchange knowledge packets that encapsulate structured concept embeddings, abstraction graphs, reasoning traces, and provenance metadata. Unlike classical distillation, M2KT operates primarily in concept space rather than example space, and it does not require labeled datasets or teacher-generated outputs during transfer. We formalize the notion of concept manifolds, introduce an inter-model alignment mapping between teacher and student latent spaces, and derive a composite loss that enforces geometric, structural, and reasoning consistency together with explicit safety constraints. We further present algorithmic procedures for teacher-side packet generation and student-side ingestion and verification. Experiments on symbolic reasoning with large language models show that M2KT can achieve approximately 85 to 90 percent of teacher performance while reducing data usage by over 98 percent compared to standard knowledge distillation. This work establishes a theoretical and practical foundation for data-free AI-to-AI knowledge transfer and self-improving model ecosystems.

تم تقديم إطار جديد يسمى نقل المعرفة من نموذج إلى نموذج (M2KT) يسمح للشبكات العصبية بنقل المعرفة دون الاعتماد على مجموعات بيانات كبيرة. تتيح هذه الطريقة الخالية من البيانات للنماذج تبادل تجسيدات المفاهيم المنظمة وآثار التفكير، مما يمثل تحولًا كبيرًا عن الأساليب التقليدية المعتمدة على البيانات مثل تقطير المعرفة والتعلم بالنقل.

Se ha introducido un nuevo marco llamado Transmisión de Conocimiento de Modelo a Modelo (M2KT), que permite a las redes neuronales transferir conocimiento sin depender de grandes conjuntos de datos. Este enfoque libre de datos permite a los modelos intercambiar embeddings de conceptos estructurados y trazas de razonamiento, marcando un cambio significativo respecto a los métodos tradicionales impulsados por datos como la destilación de conocimiento y el aprendizaje por transferencia.

Un nouveau cadre appelé Transmission de Connaissances de Modèle à Modèle (M2KT) a été introduit, permettant aux réseaux neuronaux de transférer des connaissances sans dépendre de grands ensembles de données. Cette approche sans données permet aux modèles d'échanger des embeddings de concepts structurés et des traces de raisonnement, marquant un changement significatif par rapport aux méthodes traditionnelles basées sur les données telles que la distillation de connaissances et l'apprentissage par transfert.

A new framework called Model-to-Model Knowledge Transmission (M2KT) has been introduced, allowing neural networks to transfer knowledge without relying on large datasets. This data-free approach enables models to exchange structured concept embeddings and reasoning traces, marking a significant shift from traditional data-driven methods like knowledge distillation and transfer learning.

Model-to-Model Knowledge Transmission (M2KT): A Data-Free Framework for Cross-Model Understanding Transfer

arXiv:2511.19265v1 Announce Type: new 
Abstract: The black box nature of deep neural networks poses a significant challenge for the deployment of transparent and trustworthy artificial intelligence (AI) systems. With the growing presence of AI in society, it becomes increasingly important to develop methods that can explain and interpret the decisions made by these systems. To address this, mechanistic interpretability (MI) emerged as a promising and distinctive research program within the broader field of explainable artificial intelligence (XAI). MI is the process of studying the inner computations of neural networks and translating them into human-understandable algorithms. It encompasses reverse engineering techniques aimed at uncovering the computational algorithms implemented by neural networks. In this article, we propose a unified taxonomy of MI approaches and provide a detailed analysis of key techniques, illustrated with concrete examples and pseudo-code. We contextualize MI within the broader interpretability landscape, comparing its goals, methods, and insights to other strands of XAI. Additionally, we trace the development of MI as a research area, highlighting its conceptual roots and the accelerating pace of recent work. We argue that MI holds significant potential to support a more scientific understanding of machine learning systems -- treating models not only as tools for solving tasks, but also as systems to be studied and understood. We hope to invite new researchers into the field of mechanistic interpretability.

تسلط دراسة جديدة الضوء على أهمية التفسير الميكانيكي (MI) في فهم عمليات اتخاذ القرار في الشبكات العصبية العميقة، حيث تتناول التحديات التي تطرحها طبيعتها كصندوق أسود. تقترح هذه الأبحاث تصنيفًا موحدًا لأساليب MI، مما يوفر رؤى حول كيفية عمل الشبكات العصبية وترجمتها إلى خوارزميات مفهومة.

Un nuevo estudio destaca la importancia de la interpretabilidad mecanicista (IM) en la comprensión de los procesos de toma de decisiones de las redes neuronales profundas, abordando los desafíos que plantea su naturaleza de caja negra. Esta investigación propone una taxonomía unificada de enfoques de IM, ofreciendo información sobre el funcionamiento interno de las redes neuronales y traduciéndolas en algoritmos comprensibles.

Une nouvelle étude souligne l'importance de l'interprétabilité mécaniste (IM) pour comprendre les processus décisionnels des réseaux de neurones profonds, en abordant les défis posés par leur nature de boîte noire. Cette recherche propose une taxonomie unifiée des approches d'IM, offrant des aperçus sur le fonctionnement interne des réseaux de neurones et les traduisant en algorithmes compréhensibles.

A new study highlights the importance of mechanistic interpretability (MI) in understanding the decision-making processes of deep neural networks, addressing the challenges posed by their black box nature. This research proposes a unified taxonomy of MI approaches, offering insights into the inner workings of neural networks and translating them into comprehensible algorithms.

MiniFool - Physics-Constraint-Aware Minimizer-Based Adversarial Attacks in Deep Neural Networks

Was this article worth reading? Share it

Moadly

Octofy

Dyad