ON SOME ENTROPY METHODS IN DATA-ANALYSIS

We give a report and some generalizations of two methods of data analy sis which where recently developed. The methods are based on entropies or entropy-like quantities called 'generalized mutual information' an d 'entropy profile'. They measure, each in its specific way, statistic al dependences in a time series or in a pattern. The quantities are in variant with respect to non-linear distortions of the data. All method s are, in principle, applicable to continuous stationary time series, or, more generally, to any continuous random vector. However, in pract ice also discrete processes can be investigated, provided the random v ariables attain a multitude of different values. For each method we ha ve obvious numerical algorithms.