On the recognition of time-varying patterns using learning procedures

S. Tamura,S. Higuchi,Kokichi Tanaka
DOI: https://doi.org/10.1109/TIT.1971.1054664
IF: 2.5
1971-07-01
IEEE Transactions on Information Theory
Abstract:Some recognizers for stochastic time-varying patterns with additive noise are studied. As in binary communication channels with fading, it is supposed that the fluctuation of a pattern (or signal) may be approximated by a stationary Gaussian autoregressive process with known parameters. Each measurement belongs to either of two classes: the pattern plus noise or noise alone. Under these assumptions, optimum dichotomizers with supervized learning are discussed. To the nonsupervised problems, the decision-directed approach and the modified-decision-directed approach are applied. Also some experimental results are presented.
What problem does this paper attempt to address?