STARS: Spike Tail-Aware Relational Synthesis for ANN-to-SNN Data-Free Knowledge Distillation
A new method called Spike Tail-Aware Relational Synthesis (STARS) has been proposed for ANN-to-SNN data-free knowledge distillation. STARS augments standard BN-guided synthesis with Relational Consistency Alignment and Tail-Aware Regularization to improve SNN student performance. The method is plug-and-play and addresses under-constrained issues in existing DFKD approaches. The paper is available on arXiv with ID 2605.27409.
Key facts
- STARS is a plug-and-play method for ANN-to-SNN DFKD.
- It augments BN-guided synthesis with Relational Consistency Alignment and Tail-Aware Regularization.
- Relational Consistency Alignment preserves cross-sample relational consistency between teacher and student.
- Tail-Aware Regularization regularizes threshold-relevant dynamics.
- The method targets under-constrained issues in existing DFKD approaches.
- The paper is published on arXiv with ID 2605.27409.
- The announcement type is cross.
- The method aims to improve SNN student performance in data-free settings.
Entities
Institutions
- arXiv