Ensemble estimators for accurate estimation of information measures
Information Measures in Data-driven systems In the last decade, data driven systems and algorithms have become extremely popular in a variety of fields. Examples include computer vision, machine learning, security and surveillance, artificial intelligence, transportation systems, bio-informatics, epidemiology, environmental monitoring, trading and others. Information measures (such as mutual information, entropy and divergence, minimum volume sets and so on), which characterize and help contrast this data, are a crucial part of such data-driven systems. For successful deployment of these systems, it is critical that these information measures are accurately estimated.
Ensemble Estimators of Information Measures A research group in Electrical Engineering and Computer Science of University of Michigan has reported a new method for information estimation having low implementation complexity and unprecedented accuracy. The new method, called ensemble estimator, is a versatile procedure that boosts the performance of a collection of existing information estimation methods. The estimation accuracy of the proposed method is significantly improved by a factor of 10 at the very least, which is validated by both analytical proofs and numerical simulations. Moreover, the implementation of the proposed method requires little extra effort as it builds on existing estimation methods. Therefore, the invention can be readily implemented in existing systems, which shall translate into low processing costs and fast runtime.
Application • Information estimation in data driven systems.
Advantages • Much better estimation accuracy than existing methods. • Easy implementation in existing systems. • Low processing costs and fast runtime.