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

Mohammad Ali Vahedifar, Abhisek Ray, Qi Zhang · 2026

Read Shapley Neuron Values for Continual Learning: Which Neurons Matter Most? by Mohammad Ali Vahedifar, Abhisek Ray, Qi Zhang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-15877.txt.

Description

Mohammad Ali Vahedifar, Abhisek Ray and Qi Zhang address catastrophic forgetting in continual learning with Shapley Neuron Valuation. The tool scores each neuron's value through cooperative game theory. It freezes key neurons while leaving others plastic, enabling buffer-free ongoing learning with no growth in architecture. On ImageNet-1k, it always beats rival buffer-free methods, improving accuracy by 2.88 percent in class-by-class learning and by 6.46 percent in task-by-task learning over the second baseline.

Catalog record

CREATORMohammad Ali Vahedifar, Abhisek Ray, Qi Zhang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE76233
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.15877

Browse the full library · About the Tower of Babel