Chain-Rule Processor: Optimal Classification Through Signal Processing

摘要

The chain-rule processor is a method of constructing an optimal Bayes classifier from a bank of processors. Each processor is a feature extractor designed to separate the given class from a class-dependent reference hypothesis, thereby avoiding the curse of dimensionality. This work builds upon prior work in optimal classifier design using class-specific features. The chain-rule processor is an improvement that recursively applies the PDF projection theorem.