A New Algorithm of ICA: Using the Parametrized Orthogonal Matrixes of Any Dimensions
-
摘要: 在独立分量分析 (Independent component analysis, ICA) 中, 寻找去除高阶相关的正交矩阵成为问题关键, 而正交矩阵具有特殊的空间结构, 组成它的每个列向量可视作 RN 中单位超球表面上一点, 当这些点彼此垂直时, 整体就组成一个正交矩阵. 自然地, 这些点可以用其球坐标来参数化. 本文通过观察正交矩阵的几何结构, 找到了任意维数的随机正交矩阵的参数表示方法, 且论证了这种表示的完备性; 同时, 对随机正交矩阵参数表示的随机性做了定量分析; 然后, 利用遗传算法对参数化正交矩阵中的参数进行搜索, 得到了分离结果. 本文称这种算法为 OICA 算法, 并给出了该算法的仿真实验.Abstract: In independent component analysis (ICA) problems, it is the key to find an orthogonal matrix to throw away high-order redundancy between components. Orthogonal matrixes have special structures in which row vectors of them can be seen as vectors perpendicular to each other on N-sphere. A parametrization method of all the matrices of any dimensions is proposed and the completeness of the parametric presentation is proven. Further, a theoretical analysis is made for randomicity. We use genetic algorithm (GA) in searching of the optimized separating matrix. Finally, some numerical simulation results are given.
-
Key words:
- Independent component analysis(ICA) /
- orthogonal matrix /
- random matrix /
- N-sphere /
- genetic algorithm(GA) /
- OICA
点击查看大图
计量
- 文章访问数: 2981
- HTML全文浏览量: 93
- PDF下载量: 1557
- 被引次数: 0