The purpose of data description is to give a compact description of the target data that represents most of its characteristics. In a support vector data description (SVDD), the compact description of target data is given in a hyperspherical model, which is determined by a small portion of data called support vectors. Despite the usefulness of the conventional SVDD, however, it may not identify the optimal solution of target description especially when the support vectors do not have the overall characteristics of the target data. To address the issue in SVDD methodology, we propose a new SVDD by introducing new distance measurements based on the notion of a relative density degree for each data point in order to reflect the distribution of a given data set. Moreover, for a real application, we extend the proposed method for the protein localization prediction problem which is a multiclass and multilabel problem. Experiments with various real data sets show promising results.

We propose a speed estimation method with human body accelerations measured on the chest by a tri-axial accelerometer. To estimate the speed we segmented the acceleration signal into strides measuring stride time, and applied two neural networks into the patterns parameterized from each stride calculating stride length. The first neural network determines whether the subject walks or runs, and the second neural network with different node interactions according to the subject's status estimates stride length. Walking or running speed is calculated with the estimated stride length divided by the measured stride time. The neural networks were trained by patterns obtained from 15 subjects and then validated by 2 untrained subjects' patterns. The result shows good agreement between actual and estimated speeds presenting the linear correlation coefficient r=0.9874. We also applied the method to the real field and track data.

No abstract

No abstract