CATFormer: When Continual Learning Meets Spiking Transformers With Dynamic Thresholds
CATFormer:当持续学习与具有动态阈值的脉冲变换器相遇
机构 * SustainAI Lab, MFSDS&AI(可持续人工智能实验室,多模态智能系统与人工智能)
AI总结 CATFormer通过动态阈值漏积分-放电神经元模型和门控动态头部选择机制,解决了持续学习中的灾难性遗忘问题,实现了高效能的类增量学习。
Comments Accepted for publication in the proceedings of the Neuro for AI & AI for Neuro Workshop at AAAI 2026 (PMLR)