An Evolutionary Multitasking Algorithm with Multiple Filtering for High-Dimensional Feature Selection
Contribution
Scales feature selection with multitasking and filtering, making evolutionary search workable on high dimensions.
Abstract
Many real-world feature selection problems involve hundreds of thousands of features, far beyond the scale where standard evolutionary algorithms can search effectively. This paper proposes an evolutionary multitasking algorithm with multiple filtering for high-dimensional feature selection: it treats related feature-selection subproblems as parallel tasks, transfers useful structure across them, and uses a hierarchy of cheap filters to prune the search space before invoking expensive evaluators. The result is competitive accuracy at a fraction of the runtime of competing evolutionary and non-evolutionary baselines on ultra-high-dimensional benchmarks.
Why This Paper Matters
A practical recipe for scaling evolutionary feature selection to problems where it previously could not run, with a clean multitasking formulation that generalizes to other large-scale sparse optimization tasks.
BibTeX
@article{li2023evolutionary,
title={An Evolutionary Multitasking Algorithm with Multiple Filtering for High-Dimensional Feature Selection},
author={Li, Lingjie and Xuan, Manlin and Lin, Qiuzhen and Jiang, Min and Ming, Zhong and Tan, Kay Chen},
journal={IEEE Transactions on Evolutionary Computation},
year={2023},
publisher={IEEE},
}