arXiv:2511.12095v2 Announce Type: replace 
Abstract: Event cameras sense brightness changes and output binary asynchronous event streams, attracting increasing attention. Their bio-inspired dynamics align well with spiking neural networks (SNNs), offering a promising energy-efficient alternative to conventional vision systems. However, SNNs remain costly to train due to temporal coding, which limits their practical deployment. To alleviate the high training cost of SNNs, we introduce \textbf{PACE} (Phase-Aligned Condensation for Events), the first dataset distillation framework to SNNs and event-based vision. PACE distills a large training dataset into a compact synthetic one that enables fast SNN training, which is achieved by two core modules: \textbf{ST-DSM} and \textbf{PEQ-N}. ST-DSM uses residual membrane potentials to densify spike-based features (SDR) and to perform fine-grained spatiotemporal matching of amplitude and phase (ST-SM), while PEQ-N provides a plug-and-play straight through probabilistic integer quantizer compatible with standard event-frame pipelines. Across DVS-Gesture, CIFAR10-DVS, and N-MNIST datasets, PACE outperforms existing coreset selection and dataset distillation baselines, with particularly strong gains on dynamic event streams and at low or moderate IPC. Specifically, on N-MNIST, it achieves \(84.4\%\) accuracy, about \(85\%\) of the full training set performance, while reducing training time by more than \(50\times\) and storage cost by \(6000\times\), yielding compact surrogates that enable minute-scale SNN training and efficient edge deployment.

يتناول المقال تطوير PACE (تكثيف متزامن للأحداث)، وهو إطار جديد لتقطير مجموعات البيانات يهدف إلى تحسين كفاءة تدريب الشبكات العصبية النابضة (SNN) باستخدام كاميرات الأحداث. من خلال تكثيف مجموعات بيانات كبيرة إلى نسخ اصطناعية مضغوطة، يسهل PACE تدريبًا أسرع من خلال وحدتين رئيسيتين: ST-DSM و PEQ-N. تعالج هذه الطريقة التكاليف العالية للتدريب المرتبطة بـ SNN، والتي كانت عقبة أمام تطبيقها العملي في أنظمة الرؤية المعتمدة على الأحداث.

El artículo aborda el desarrollo de PACE (Condensación Fase-Alineada para Eventos), un nuevo marco de destilación de conjuntos de datos destinado a mejorar la eficiencia del entrenamiento de redes neuronales de picos (SNN) utilizando cámaras de eventos. Al condensar grandes conjuntos de datos en versiones sintéticas compactas, PACE facilita un entrenamiento más rápido a través de dos módulos clave: ST-DSM y PEQ-N. Este enfoque aborda los altos costos de entrenamiento asociados con los SNN, que han sido una barrera para su aplicación práctica en sistemas de visión basados en eventos.

L'article traite du développement de PACE (Phase-Aligned Condensation for Events), un nouveau cadre de distillation de jeux de données visant à améliorer l'efficacité de l'entraînement des réseaux de neurones à impulsions (SNN) utilisant des caméras événementielles. En condensant de grands ensembles de données en versions synthétiques compactes, PACE facilite un entraînement plus rapide grâce à deux modules clés : ST-DSM et PEQ-N. Cette approche répond aux coûts d'entraînement élevés associés aux SNN, qui ont été un obstacle à leur application pratique dans les systèmes de vision basés sur des…

The article discusses the development of PACE (Phase-Aligned Condensation for Events), a novel dataset distillation framework aimed at enhancing the training efficiency of spiking neural networks (SNNs) using event cameras. By condensing large datasets into compact synthetic versions, PACE facilitates faster training through two key modules: ST-DSM and PEQ-N. This approach addresses the high training costs associated with SNNs, which have been a barrier to their practical application in event-based vision systems.

Learning from Dense Events: Towards Fast Spiking Neural Networks Training via Event Dataset Distillation

Was this article worth reading? Share it

Dynamiq

ChartPixel

pikaai.org