Dynamic Multitask Learning Framework for High-Dimensional Feature Selection

A new research study has shed light on the challenges of high-dimensional data and proposed a dynamic multitask learning framework to address these challenges. The framework, designed by researchers at China Tobacco Zhejiang Industrial Co. Ltd., is an evolutionary optimization setting that integrates competitive learning and knowledge transfer. The method combines multiple feature relevance indicators to generate two complementary tasks, which are optimized in parallel using a competitive particle swarm optimization algorithm enhanced with hierarchical elite learning. This approach is shown to be highly effective in balancing exploration, exploitation, and knowledge sharing for robust feature selection.

Key Takeaways:

• The proposed dynamic multitask learning framework is designed to address the challenges of high-dimensional data, which often contain noisy and redundant features.
• The framework generates two complementary tasks through a multi-criteria strategy that combines multiple feature relevance indicators, ensuring both global comprehensiveness and local focus.
• The method optimizes these tasks in parallel using a competitive particle swarm optimization algorithm enhanced with hierarchical elite learning.
• The approach is shown to be highly effective in balancing exploration, exploitation, and knowledge sharing for robust feature selection, achieving superior classification accuracy with fewer selected features.
• The researchers claim that the proposed method achieves the highest accuracy on 11 out of 13 benchmark datasets and the fewest features on eight out of 13 datasets.
• The study reports an average accuracy of 87.24% and an average dimensionality reduction of 96.2% across the 13 benchmark datasets.

Statistics:

• 13 high-dimensional benchmark datasets were used to test the proposed method.
• The average accuracy achieved by the proposed method is 87.24%.
• The average dimensionality reduction achieved by the proposed method is 96.2% (median 200 selected features).
• The proposed method achieves the highest accuracy on 11 out of 13 benchmark datasets and the fewest features on eight out of 13 datasets.

Sources:

• NewsRx. Study Findings on Artificial Intelligence Reported by Researchers at China Tobacco Zhejiang Industrial Co. Ltd. (A dynamic multitask evolutionary algorithm for high-dimensional feature selection based on multi-indicator task construction and elite competition learning). Robotics & Machine Learning. November 3, 2025; p 488.
• Frontiers in Artificial Intelligence. A dynamic multitask evolutionary algorithm for high-dimensional feature selection based on multi-indicator task construction and elite competition learning, doi: 10.3389/frai.2025.1667167.
• Frontiers Media S.A., publisher of Frontiers in Artificial Intelligence.