Shapley Neuron Values for Continual Learning: Which Neurons Matter Most?
The paper titled 'Shapley Neuron Values for Continual Learning: Which Neurons Matter Most?' introduces a framework for addressing catastrophic forgetting in neural networks. The Shapley Neuron Valuation (SNV) method quantifies neuron importance and allows for buffer-free continual learning. Experimental results demonstrate that SNV outperforms existing methods, improving accuracy significantly in various learning scenarios.
- ▪The paper presents a framework called Shapley Neuron Valuation (SNV) to tackle catastrophic forgetting in neural networks.
- ▪SNV selectively freezes important neurons while keeping others plastic, facilitating continual learning without expanding the architecture.
- ▪Experiments on ImageNet-1k show that SNV improves accuracy by +2.88% in class incremental learning and +6.46% in task incremental learning compared to the second baseline.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Machine Learning arXiv:2605.15877 (cs) [Submitted on 15 May 2026] Title:Shapley Neuron Values for Continual Learning: Which Neurons Matter Most? Authors:Mohammad Ali Vahedifar, Abhisek Ray, Qi Zhang View a PDF of the paper titled Shapley Neuron Values for Continual Learning: Which Neurons Matter Most?, by Mohammad Ali Vahedifar and Abhisek Ray and Qi Zhang View PDF HTML (experimental) Abstract: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.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
