Related Experiment Video
Updated: Aug 5, 2026
07:35
Selecting Multiple Biomarker Subsets with Similarly Effective Binary Classification Performances
Published on: October 11, 2018
From Global to Local: Semantic-Aware Instance-Wise Feature Selection
Zihan Wang1, Yue Zhang2, Hengpeng Xu3
1College of Mathematics and Statistics Science, Ludong University, Yantai 264000, China.
Entropy (Basel, Switzerland)
|July 28, 2026
Summary
This study introduces Semantic-aware Instance-wise Feature selection (SIF), a new method for dimension reduction. SIF enhances feature selection by considering both semantic correlations and instance-specific characteristics for improved performance.
Area of Science:
- Computer Science
- Artificial Intelligence
- Machine Learning
Background:
- Existing feature selection methods often focus on global or local criteria alone.
- Few approaches consider feature selection granularity beyond the instance level.
- Assessing feature significance from an individual view presents limitations.
Purpose of the Study:
- To present a novel Semantic-aware Instance-wise Feature selection (SIF) model.
- To address the weaknesses of existing feature selection methods.
- To specify feature representations at the instance level, overcoming learning complexity.
Main Methods:
- SIF employs a sequential pipeline framework.
- It models semantic correlations to select semantic-aware features.
- Inconsistent instances are captured to guide instance-wise feature selection.
Main Results:
- The final optimal feature subset combines semantic-aware and instance-wise features.
- SIF represents semantics globally and describes instance characteristics locally.
- Extensive experiments demonstrate SIF's superiority across various metrics.
Conclusions:
- SIF offers a holistic approach to feature selection.
- The model effectively integrates global semantic information and local instance details.
- SIF advances dimension reduction techniques by considering instance-level feature representation.
