The aim of this study was to improve the robust weighted averaging based on criterion function minimization and assess its effectiveness for extracting event-related brain potentials (ERP) from electroencephalographic (EEG) recordings. The areas of improvement include significantly lower averaging error (45% lower RMSE and 37% lower maximum difference than for original implementation) and increased robustness to local minima, strong outliers and corrupted epochs common to real-life EEG signals, especially from low-cost devices. Our proposed procedure was tested on two datasets, one artificially generated for purposes of this study (including different noise sources) and one real-life dataset collected with Emotiv EPOCþ. The lower error results mainly from more effective rejection (lowering the weights) of corrupted epochs by integrating the correlation-based weighting. The advantages of our method over pure correlation-based weighting are lower RMSE (up to two times) and robustness to the algorithm initialization and strong outliers. The performance of the methods was measured using bootstrap testing to avoid dependency of results on data. It shows that our improvements lead to significantly lower error, especially when the EEG signal is not filtered. The values of the parameters were adjusted for EEG signals but they can easily be incorporated in other repetitive electrophysiological measurement techniques.
2
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
Na przykładzie międzylaboratoryjnych badań porównawczych precyzji (niepewności) pewnej metody pomiarowej, omówiono jak na dokładność oceny statystycznej ich wyników wpływają outliery (dane odstające), jeśli pojawią się w tych wynikach. Rozpatrzono możliwość zastosowania odpornych metod oszacowania jako alternatywę do tradycyjnie stosowanego odrzucania danych odstających. Uwzględniają one wyniki wszystkich pomiarów wraz z outlierami. Pozwalają też na bardziej wiarygodne statystycznie oszacowanie rozkładu normalnego modelującego dane eksperymentalne, szczególnie dla małych próbek. Jako ilustrację, oszacowano wspólne odchylenie standardowe precyzji pewnej metody pomiarowej dla wyników badań tą metodą otrzymanych w 9-ciu laboratoriach. Odchylenie to, obliczone tradycyjnie bez odrzucenia outliera, było 1,5 razy większe niż z odrzuceniem, zaś dla metody odpornej "S- algorytm" jest bliskie mniejszej z obu wartości, lecz ma większą od niego wiarygodność.
EN
The influence of outliers in measurement results on the accuracy of resulting estimates is shown. Implementation of robust estimation methods is considered. These methods take into account all measurement results including outliers and the corresponding to them normal distribution could be choose better. Then it allows to provide a more reliable statistical estimates than classic methods with eliminating outliers, especially for samples of small volume. As the example the estimates of the common standard deviation of all numerical data from comparing tests of the measurement precision of some method in 9 labs are calculated by traditional methods and robust method "S-Algorithm". Results confirm the better efficiency of this robust method.
3
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
This paper introduces a new classifier design method based on a kernel extension of the classical Ho-Kashyap procedure. The proposed method uses an approximation of the absolute error rather than the squared error to design a classifier, which leads to robustness against outliers and a better approximation of the misclassification error. Additionally, easy control of the generalization ability is obtained using the structural risk minimization induction principle from statistical learning theory. Finally, examples are given to demonstrate the validity of the introduced method.
4
Dostęp do pełnego tekstu na zewnętrznej witrynie WWW
A new learning method tolerant of imprecision is introduced and used in neuro-fuzzy modelling. The proposed method makes it possible to dispose of an intrinsic inconsistency of neuro-fuzzy modelling, where zero-tolerance learning is used to obtain a fuzzy model tolerant of imprecision. This new method can be called e-insensitive learning, where, in order to fit the fuzzy model to real data, the e-insensitive loss function is used. e-insensitive learning leads to a model with minimal Vapnik-Chervonenkis dimension, which results in an improved generalization ability of this system. Another advantage of the proposed method is its robustness against outliers. This paper introduces two approaches to solving e-insensitive learning problem. The first approach leads to a quadratic programming problem with bound constraints and one linear equality constraint. The second approach leads to a problem of solving a system of linear inequalities. Two computationally efficient numerical methods for e-insensitive learning are proposed. Finally, examples are given to demonstrate the validity of the introduced methods.
JavaScript jest wyłączony w Twojej przeglądarce internetowej. Włącz go, a następnie odśwież stronę, aby móc w pełni z niej korzystać.