Neural ISOMAP
Shih-Pin Chao, Chen-Lan Yen, Chien-Chun Kuo · 2007
In recent years, the studies of digital content engineering confront us with massive amounts of data for classification and analysis, such as, thousands of news videos, surveillance records, motion capture data, images of animals and plants, etc. For these studies, the relationships between each data point are often hidden in a multi-dimensional space. For the reveal of the relationships between each data point, the ISOMAP method is often used. This is because that ISOMAP preserves the intrinsic dimensionality and metric structure of data. Therefore, this paper proposes a neural network-based ISOMAP method to efficiently obtain an ISOMAP robustly and stable. The benefits of the proposed method are that the time complexity is linear and space complexity is constant.