Inspired by the advantages of the matrix learning and the SSL, in this paper, we propose a novel semi-supervised matrix learning algorithm by incorporating the manifold regularization into the matrixized least squares support vector machine (MatLSSVM), termed as Laplacian matrixized LSSVM, or LapMatLSSVM for short.
Thus, prediction of the 1994 BLC6 vectors from their previous 5BLC, by using the supervised W matrix with 1992_93, gave an average <δ_pred> = 1.25, which decreases to 1.11 by smoothing the matrix indicating that it represents better the 1994 test period.
In this study, OLPP is applied with a supervised similarity matrix for comparison.
Meanwhile, it makes the cross-domain gaits for different subjects disperse more separately with a large margin by using the supervised similarity matrix.
As feature extraction based on supervised Nonnegative Matrix Factorization (NMF) has been proposed in automatic speech recognition for enhanced robustness, we introduce and evaluate different kinds of NMF-based features for emotion recognition.
The supervised trained matrix is then used for a real BLC6 prediction from their previous 5BLC, but in a test period (1995) not used in the training.
Hence, in order to analyze the most probable sequences of five winds preceding a particular class of wind for the 6th day, we have used the supervised trained matrix (smoothed 3 times) and its 162×162 Kohonen map, with 1975 94 wind data and 10000 epochs, which corresponds to the neighbourhood matrix of Figure 4.
β is used to prevent the supervised kernel distance matrix KDist from increasing too fast when the kernel Euclidean distance matrix Dist is relatively large, since Dist is in the exponent.
where KDist is the supervised kernel distance matrix with the class label information, while Dist is the kernel Euclidean distance matrix without the class label information.
The supervised-trained 3D matrix and its correspondent Kohonen map, give then information about the most probable CWT sequences before each one of the 27 CWT's.
