The present invention performs linear transformation on the normal coordinate system of the positive definite matrix (SPD matrix) to remove nonlinear constraints to improve speed and accuracy during training. A log-Euclidean metric learning apparatus and method using a sub-manifold framework, comprising: a tangent space mapping unit that maps data expressed in a symmetric positive definition (SPD) matrix to a tangent space; tangent space mapping A Euclidean point processing unit that represents points mapped from the sub-points as Euclidean points (R D ); A subspace mapping unit that maps to a subspace of (R K) through a parameter W; Remap from the tangent space to SPD(n) through expm (matrix exponential), and reduce the distance for points of the same class and reduce the distance for points of the other class using an objective function. It includes; a remapping unit that increases and performs metric learning.
展开▼
机译:本发明对正定矩阵(SPD矩阵)的正常坐标系上进行线性变换,以去除非线性约束,以提高训练期间的速度和精度。使用子歧管框架的日志欧几里德度量学习装置和方法,包括:切线空间映射单元,其映射在对称的正定义(SPD)矩阵中以对称的空间表示的数据;切线空间映射欧几里德点处理单元,其表示从子点映射的点作为欧几里德点(R d sup>);子空间映射单元通过参数w映射到(rk)的子空间; sup>从切换空间汇到spd(n)通过扩张(矩阵指数),并减少点的距离使用目标函数相同的类并减少其他类的点的距离。这包括;一种增加和执行度量学习的重新映射单元。
展开▼