Representation of Spatio-tempral Patterns with Nonlinear Manifold and its Application to Gestrue Recognition

时空模式被认为是羽毛空间中的一组特征向量，称为流形。在该研究中，基于图距离的自组织映射(GSOM)充分逼近了流形，而且流形之间的相似性代表了特殊的形式，高阶SOM的学习往往变得不稳定。改进了学习算法，实现了稳定的学习。

Spatio-temporal patterns is considered as a set of feature vectors in a feaure space called as a manifold . In this study，the manifolds were adequately approximated using graph distance based self-organizing map (GSOM)，furthermore，similarity between manifolds represent peculiar forms，and the learning of higher rank SOM often becomes unstable. The learning algorithm was modified to realize a stable learning.