The support vector machine (SVM) is a method for classification and for function approximation. This method commonly makes use of an /spl epsi/-insensitive cost function, meaning that errors smaller than /spl epsi/ remain unpunished. As an alternative, a least squares support vector machine (LSSVM) uses a quadratic cost function. When the LSSVM method is used for function approximation, a nonsparse solution is obtained. The sparseness is imposed by pruning, i.e., recursively solving the approximation problem and subsequently omitting data that has a small error in the previous pass. However, omitting data with a small approximation error in the previous pass does not reliably predict what the error will be after the sample has been omitted. In this paper, a procedure is introduced that selects from a data set the training sample that will introduce the smallest approximation error when it will be omitted. It is shown that this pruning scheme outperforms the standard one.

No abstract

This note describes the design and testing of a programmable pulsatile flow pump using an Arduino micro-controller. The goal of this work is to build a compact and affordable system that can relatively easily be programmed to generate physiological waveforms. The system described here was designed to be used in an in-vitro set-up for vascular access hemodynamics research, and hence incorporates a gear pump that delivers a mean flow of 900 ml/min in a test flow loop, and a peak flow of 1106 ml/min. After a number of simple identification experiments to assess the dynamic behaviour of the system, a feed-forward control routine was implemented. The resulting system was shown to be able to produce the targeted representative waveform with less than 3.6% error. Finally, we outline how to further increase the accuracy of the system, and how to adapt it to specific user needs.

The power spectrum of an EEG signal shows differences with respect to its baseline the moment a subject is hearing, or expecting, a tone. As this difference also occurs when one is not actually hearing it, a Brain Computer Interface can be developed in which imagined rhythms are used to transfer information. Four healthy subjects participated in this study in which they had to imagine a simple rhythm. A metronome was kept ticking so that the subjects would not drift in their tempo. Solely based on the EEG signals, the classifier had to distinguish between imagined accented and non-accented tones. The features for the classification were automatically selected out of a set of possible features that focussed on phase and power differences of independent components. The classification rate found is about 0.6 for two of the four subjects, and several classifications can be combined to increase this classification rate to values larger than 0.7 with 2 s worth of data for the best performing subject. Chance level for our classification task is 0.5.