Related Experiment Video
Updated: May 28, 2026
07:35
Selecting Multiple Biomarker Subsets with Similarly Effective Binary Classification Performances
Published on: October 11, 2018
Federated Multi-View Unsupervised Feature Selection via Bio-Inspired Hierarchical-Cognitive Tianji's Horse Racing
Rong Cheng1, Zhiwei Sun2, Kun Qi1
1School of Electronic and Communication Engineering, Shenzhen Polytechnic University, Shenzhen 518055, China.
Biomimetics (Basel, Switzerland)
|May 26, 2026
Summary
This study introduces Fed-MUFSHT, a federated learning framework for multi-view unsupervised feature selection (MUFS). It enhances optimization and convergence for privacy-preserving distributed machine learning tasks.
Area of Science:
- Machine Learning
- Data Science
- Artificial Intelligence
Background:
- Multi-view datasets are prevalent but often unlabeled and distributed, challenging traditional centralized machine learning.
- Existing federated and centralized methods struggle with local optima and convergence guarantees in federated multi-view feature selection.
- Privacy concerns and communication constraints hinder the application of centralized methods to distributed multi-view data.
Purpose of the Study:
- To propose Fed-MUFSHT, a federated framework for multi-view unsupervised feature selection (MUFS).
- To address challenges of privacy, communication, local optima, and convergence in distributed multi-view data.
- To enhance feature selection performance and robustness in federated learning settings.
Main Methods:
- Fed-MUFSHT integrates tensor learning (TL) with a novel metaheuristic optimizer, Hierarchical-Cognitive Tianji's Horse Racing Optimization (HC-THRO).
- A dual-stage local optimization process includes HC-THRO (Hierarchical Competitive Learning and Adaptive Cognitive Mapping) for exploration and Stage 2 TL for imputation and representation.
- Global model coordination uses a privacy-preserving aggregation strategy based on Normalized Mutual Information (NMI) and feature weights.
Main Results:
- Fed-MUFSHT demonstrated clear advantages over competing methods on benchmark datasets.
- The proposed framework achieved better optimization results and more dependable convergence characteristics.
- Experiments confirmed the robustness and effectiveness of Fed-MUFSHT for distributed optimization with privacy protection.
Conclusions:
- Fed-MUFSHT offers a robust and effective solution for multi-view unsupervised feature selection in federated learning.
- The integration of tensor learning and HC-THRO significantly improves optimization and convergence.
- The framework successfully addresses privacy and communication constraints in distributed multi-view data analysis.
