One of the main obstacles to the useful exploitation of dimensionality reduction, both linear and nonlinear, is the lack of effective synthesis methods to generate examples in the original exploitation space, as opposed to the low dimensional parameter space. The ability to interactively navigate in the appropriate relevant low dimensional representation, and simultaneously observe the related “original” data modality would enable both sensor fusion and enhanced recognition and identification. Our team has developed an initial theoretical framework that promises to provide this ability, which we propose to demonstrate as Fast Online Interactive Generative Multiscale Manifold Learning.
Keywords: Manifold Learning, Manifold Learning, Target Recognition, Compression, Anomaly Detection, Machine Learning, Low-Dimensional, Sparse, Nonlinear
