在現實世界中,資料點往往都是在高維高間中,然而,研究人員都會去尋找在低維空間中資料點的資訊,在這篇論文中,我們採用一個新的子空間學習演算法叫做基於屬性的子空間學習。一般來說,子空間學習只考慮資料的鄰居和類的標籤來學習子空間,但我們的方法有一起考慮到資料的鄰居關係和類的標籤和資料的屬性去學習子空間,在實驗結果中,我們驗證我們的方法在戶外場景資料庫和公眾人物資料庫中做分類的時候相較於其他的子空間學習是具有競爭力的。
The data point is often in high dimensional space in real world. However , researcher seeks data points in the low dimensional space that preserved data point information. In this thesis , we propose a novel subspace learning algorithm method called Attribute based Subspace learning (ASL). Generally speaking , subspace learning only consider data neighborhood and class labels to learn the subspace. But , our method jointly consider data neighboring relation , class labels and data attributes to learn the subspace. Experimental results on OSR and PubFig datasets verification that our proposed method is very competitive to other subspace learning methods for classification.