用博弈论评估神经元重要性，缓解持续学习中的遗忘问题
Shapley Neuron Values for Continual Learning: Which Neurons Matter Most?
- 基于合作博弈论量化神经元重要性，决定哪些该冻结
- 图像分类任务中准确率提升2.88%至6.46%（无记忆缓冲区）
- 无需扩展网络结构，适合资源受限的持续学习场景
持续学习使神经网络能够顺序学习新任务而不遗忘旧知识。然而，神经网络存在灾难性遗忘问题，即学习新任务会降低对旧任务的性能。本文提出基于合作博弈论的神经元重要性评估框架——肖普利神经元值（SNV），可精确衡量持续学习中各神经元的重要性。通过选择性冻结重要神经元，同时保持其他神经元可塑性，实现无需记忆缓冲区的持续学习，且不需扩展模型架构。在ImageNet-1k上的实验表明，相比现有无缓冲区方法，SNV在类别增量学习中提升2.88%准确率，在任务增量学习中提升6.46%。
原文摘要 · Abstract (English)
Continual learning enables neural networks to learn tasks sequentially without forgetting previously acquired knowledge. However, neural networks suffer from catastrophic forgetting, where learning new tasks degrades performance on earlier ones. We address this problem with Shapley Neuron Valuation (SNV), a principled framework that quantifies Neuron importance in continual learning, grounded in cooperative game theory. SNV selectively freezes important Neurons while keeping others plastic, enabling buffer-free continual learning without expanding architecture. Experiments on ImageNet-1k show that SNV consistently outperforms existing buffer-free methods. In particular, SNV improves accuracy by +2.88% in the class incremental learning and +6.46% in the task incremental learning scenarios compared to the second baseline.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。