Research On Finsler Geometry-based Structure Learning Algorithm

Structure learning is an important method in machine learning which has received widespread attention in recent years. In allusion to questions of structural representation and structural metric, we propose Finsler Geometry-based Structure Learning Algorithm which includes the following contents:(1)Representing shapes in a simple way for classification is difficult because the set of shapes is nonlinear and infinite dimensional, hence we present Finsler Geometry Structure Learning Algorithm from the perspective of the shape’s geometric structure. This algorithm can effectively preserve the geometric structure in the process of classification.(2)A modified version based on Finsler geometry was proposed which introduces Finsler metric. This algorithm can effectively make up for the defect of similarity measure in K-means.(3)Experimental confirmation: applications of Finsler Geometric K-means in handwritten digit recognition and face recognition show the effectiveness of this algorithm.