arXiv:2605.27409v1 Announce Type: cross 
Abstract: SNNs promise energy-efficient and low-latency inference, but their performance still trails that of ANNs. ANN-to-SNN knowledge distillation helps narrow this gap, yet the original training data are often unavailable in practical deployment settings. Existing data-free knowledge distillation (DFKD) methods synthesize surrogate data by matching teacher-side priors, especially BN statistics, but these ANN-oriented constraints mainly regularize mean and variance and therefore remain under-constrained for SNN students whose responses depend on threshold-crossing dynamics. In this paper, we propose Spike Tail-Aware Relational Synthesis (STARS), a plug-and-play method for ANN-to-SNN DFKD that augments standard BN-guided synthesis with two complementary objectives: Relational Consistency Alignment, which preserves cross-sample relational consistency between teacher and student, and Tail-Aware Regularization, which regularizes threshold-relevant tail probabilities through soft exceedance over teacher-derived thresholds. Together, these objectives generate synthetic batches that remain teacher-valid while becoming more informative for SNN students. Experiments on CIFAR-10, CIFAR-100, and Tiny-ImageNet across multiple ANN-SNN pairs show that our method consistently improves conventional DFKD baselines and even surpasses several KD methods, with gains of up to 4.6\% on CIFAR-10 and 6.7\% on CIFAR-100, highlighting the importance of complementing BN matching with relational and tail-aware constraints in SNN-oriented DFKD.

قدم الباحثون طريقة جديدة تُعرف باسم التوليف العلاقي الحساس للذروات (STARS) للانتقال من الشبكات العصبية الاصطناعية (ANN) إلى الشبكات العصبية النبضية (SNN) دون الحاجة إلى بيانات. تُعزز هذه الطريقة الأساليب الحالية من خلال دمج هدفين: توافق الاتساق العلاقي والتنظيم الحساس للذيل، مما يعالج القيود المفروضة على الأساليب الحالية التي تركز بشكل أساسي على المتوسط والانحراف المعياري.

Investigadores han presentado la Síntesis Relacional Sensible a Picos (STARS), un nuevo método para la destilación de conocimiento sin datos de redes neuronales artificiales (ANN) a redes neuronales espinosas (SNN). Este enfoque mejora las técnicas existentes al incorporar dos objetivos: Alineación de Consistencia Relacional y Regularización Sensible a la Cola, abordando las limitaciones de los métodos actuales que se centran principalmente en la media y la varianza.

Des chercheurs ont introduit la Synthèse Relationnelle Sensible aux Pics (STARS), une nouvelle méthode pour la distillation de connaissances sans données des réseaux de neurones artificiels (ANN) vers les réseaux de neurones à impulsions (SNN). Cette approche améliore les techniques existantes en intégrant deux objectifs : l'Alignement de Cohérence Relationnelle et la Régularisation Sensible aux Queues, répondant aux limites des méthodes actuelles qui se concentrent principalement sur la moyenne et la variance.

Researchers have introduced Spike Tail-Aware Relational Synthesis (STARS), a novel method for data-free knowledge distillation from artificial neural networks (ANNs) to spiking neural networks (SNNs). This approach enhances existing techniques by incorporating two objectives: Relational Consistency Alignment and Tail-Aware Regularization, addressing the limitations of current data-free methods that primarily focus on mean and variance.