“AdaptCL: Data-Driven Pruning and Parameter Isolation Enhance Continual Learning on Heterogeneous Sequential Datasets”

https://pubmed.ncbi.nlm.nih.gov/38113151

The AdaptCL method, which uses fine-grained data-driven pruning and task-agnostic parameter isolation, enhances continual learning performance on heterogeneous sequential datasets by adapting to variations in data complexity, size, and similarity.