Nonlinear dimension reduction algorithm based on geodesic covering

• Main Authors: MA ZHENGMING, YIN WANGUANG, LI WEIJUN, LIU JIE, GUO JIAJING
• Format: Patent
• Language: Chinese; English
• Published: 13.07.2018
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention relates to problems related to manifold learning in machine learning, and provides a nonlinear dimension reduction algorithm based on geodesic covering. Whether the covering algorithm ofa geodesic line is good or bad not only significantly influences the calculation efficiency but also directly affects whether the dimension reduction effect is good or bad. Therefore, the invention firstly discloses a 'radial shortest path covering algorithm'. The algorithm can efficiently conduct geodesic covering on a high-dimension data sample point set, and thus the point set is integrated into a geodesic line set. Then the invention provides the nonlinear dimension reduction algorithm based on geodesic covering. The essential thought of the dimension reduction algorithm is that a prediction value exists between the points on each geodesic line and the starting point of the geodesic line, errors between the prediction value and unknown real values in the low-dimension space are accumulated, and thus according