Multilayer bootstrap networks

We describe a simple multilayer bootstrap network for unsupervised dimensionality reduction that each layer of the network is a group of mutually independent k-centers clusterings, and the centers of a clustering are randomly sampled data points. We further compress the network size of multilayer bootstrap network by a neural network in a pseudo supervised way for prediction. We report comparison results with 10 unsupervised methods in data visualization, clustering, and document retrieval.