Title :
Optimal filters for attribute generation and machine learning
Author :
Birdwell, J. Douglas ; Horn, Roger D.
Author_Institution :
Dept. of Electr. & Comput. Eng., Tennessee Univ., Knoxville, TN, USA
Abstract :
Extensions to inductive inference methods of machine learning are proposed which allow inference from dynamic information contained in sampled data signals. An optimization problem over a set of finite impulse response filters is posed which, while not convex, can provide good quality attributes for classification of signal sources. Characteristics of the optimization problem, possible methods of its solution, and results using nonlinear programming are discussed
Keywords :
digital filters; inference mechanisms; learning systems; nonlinear programming; FIR filters; attribute generation; dynamic information; finite impulse response filters; inductive inference methods; machine learning; nonlinear programming; optimization; Classification algorithms; Classification tree analysis; Data mining; Entropy; Finite impulse response filter; Machine learning; Machine learning algorithms; Optimization methods; Testing; Tree graphs;
Conference_Titel :
Decision and Control, 1990., Proceedings of the 29th IEEE Conference on
Conference_Location :
Honolulu, HI
DOI :
10.1109/CDC.1990.203869
