Abstract: Manifold learning has attracted extensive interests of researchers from machine learning and data mining. Such as, Locally Linear Embedding（LLE） algorithm has a good generalization performance as a nonlinear dimensionality reduction algorithm, and be wildly used in image classification and object recognition, but just assumes the data resides on a single manifold. Multiple Manifold Locally Linear Embedding（MM-LLE） algorithm as an improved version of considered multi-manifold, there are some shortcomings still. Therefore, an improved MM-LLE algorithm is proposed, which uses the local low dimensional manifolds between any two classes to construct classifiers to improve the classification accuracy, and the way of calculating the best dimension of the original algorithm is improved. This paper compares with algorithm ISOMAP, LLE and MM-LLE in classification accuracy. The experimental results verify the effectiveness of the proposed method.

Key words: Local Linear Embedding（LLE）, multi-manifold learning, best dimension, classification

摘要： 流形学习已成为机器学习和数据挖掘领域的研究热点。比如，算法LLE（Locally Linear Embedding）作为一种非线性降维算法有很好的泛化性能，被广泛地应用于图像分类和目标识别，但其仅仅假设了数据集处于单流形的情况。MM-LLE（Multiple Manifold Locally Linear Embedding）学习算法作为一种考虑多流形情况的改进算法，依然存在几点不足之处。因此，提出改进的MM-LLE算法，通过任意两类间的局部低维流形组合并构建分类器来提高分类精度；同时改进原算法计算最佳维度的方法。通过与算法ISOMAP、LLE以及MM-LLE比较分类精度，实验结果验证了改进算法的有效性。

关键词: 局部线性嵌入（LLE）, 多流形学习, 最佳维度, 分类