Core Concepts
Proposing a novel method, GRROOR, for multi-label feature selection using orthogonal regression to optimize global redundancy and relevance.
Abstract
新しい埋め込み型マルチラベル特徴選択手法、GRROORを提案。Least square regressionに基づく既存の手法の制限を克服し、グローバルな冗長性と関連性を最適化する。実験結果は効果的であることを示す。
Stats
Least square regression(LSR)-based multi-label feature selection methods learn a projection matrix W with sparsity restriction by minimizing regression error and the score of each feature is calculated by {∥w1∥2, ..., ∥wd∥2}.
Existing LSR-based multi-label feature selection methods have the limitation of not preserving sufficient discriminative properties in the projection subspace.
The proposed GRROOR method introduces global redundancy and relevance optimization in orthogonal regression to tackle this challenge.