Exponentially Embedded Family of Distributions for Effective Data Representation, Information Extraction, and Decision Making.

摘要

We have focused on the mathematical formalism and stochastic machine learning algorithms for extraction of relevant information based on the exponentially embedded family (EEF), learning and classification over large- scale stream data, and information fusion and integration. In particular, we have proposed a probability density function (PDF) estimation approach based on the EEF, and a measure for assessment of information from sensors. We have also taken advantage of the model structure information for model estimation. Furthermore, we have proved a general Pythagorean theorem for the EEF and studied a multi path scenario for sensor selection. Finally, we also analyzed and developed a series of machine learning techniques for effective data learning, classification, and decision making, including adaptive incremental learning from stream data, information fusion with multiple learning models/hypotheses, machine learning with non-stationary imbalanced stream data, kernel density estimation based on self-organizing map(SOM), among others. These results have been published in peer-reviewed conferences and journals, including IEEE Transactions on Neural Networks, IEEE Transactions on Neural Networks and Learning Systems, Neurocomputing (Elsevier), a book chapter with Wiley-IEEE, among others.