| 摘要 |
The invention relates to a data processing method, which runs on a data processing device, for mapping input data to be processed onto output data. According to this method: the data objects to be processed are input as input data; the input data objects are processed with aid of a topology-preserving map by the arrangement of neurons according to a predetermined schema in an arrangement space; code book objects in the result space are assigned to the neurons, and; code book objects are processed in accordance with the calculation rule of a topology-preserving map while using data objects of the investigation space. The processed code book objects are output as output data. The invention is characterized in that at least a portion of the input data objects is used in order to determine the arrangement of neurons in the arrangement space and/or in that data objects are input, which are required for data processing, are independent of the input data to be processed, and which are used as data objects of the information space. The invention additionally relates to a data processing method, which runs on a data processing device, for mapping data objects to be processed onto distance objects. According to this method: the data objects to be processed are input; data objects to be processed are input; distances between the data objects to be processed are calculated as distance objects, and; these distance objects are output as output data. This method is characterized in that the distances are calculated by means of statistical learning methods, local modules, inference statistics methods, and/or by one of the following special calculation methods: Levenstein function, mutual information, Kullback-Leibler divergence, coherence functions used in signal processing, particularly for bio-signals, LPC cepstral distance, calculating methods, that relate power spectra of two signals, such as the Itakura-Saito distance, the Mahalanobis distance, and/or calculating methods with regard to the phase synchronization of oscillators. Finally, the invention relates to a data processing method, which runs on a data processing device, for determining the cluster validity. According to this method: data objects are input; distance objects between these data objects are input and/or calculated, and; an assignment of the data objects, which are to be processed, to groups is input and/or calculated, particularly in accordance with a method as cited in Claims Nos. 1 to 5, and a measurement for the quality of this assignment is output as output data. This method is characterized in that the measurement for the quality of the assignment is calculated by using at least one portion of the input and/or calculated distance objects. The invention also relates to associated data processing devices and computer program products. |
