Shapley Neuron Values for Continual Learning: Which Neurons Matter Most?

About

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.

Mohammad Ali Vahedifar, Abhisek Ray, Qi Zhang · 2026

Related benchmarks

Task	Dataset	Result
Class-incremental learning	CIFAR100 10 Tasks	Accuracy54.7	66
Class-incremental learning	CIFAR-100 20 tasks	Accuracy44.85	58
Task-Incremental Learning	Tiny-ImageNet 20 tasks	Average Accuracy67.95	54
Task-Incremental Learning	CIFAR-100 10 tasks	Backward Transfer0	44
Task-Incremental Learning	CIFAR-100 20 tasks	Accuracy (ACC)71.93	40
Task-Incremental Learning	Tiny-ImageNet 10 tasks	Accuracy74.82	33
Class-incremental learning	Tiny-ImageNet 10 tasks	Accuracy45.7	31
Class-incremental learning	Tiny-ImageNet 20 tasks	Accuracy37.47	25
Class-incremental learning	ImageNet-1k 10 Tasks (test)	Accuracy41.3	13
Class-incremental learning	ImageNet-1k 20 Tasks (test)	Accuracy34.2	13