Dimensionality Reduction of Multimodal Labeled Data by Local Fisher Discriminant Analysis
The Journal of Machine Learning Research
Learning with l1-graph for image analysis
IEEE Transactions on Image Processing
| Hi-index | 0.10 |
An important issue in brain-computer interfaces based on multi-channel EEG signals is to extract spatial filters. Local temporal common spatial patterns (LTCSP) is a recently developed approach to find spatial filters, which takes temporally local information into account. The formulation of LTCSP, however, is essentially a subspace decomposition technique. In this paper, we extend LTCSP from the aspects of discrimination and adaption. The discriminant extension is based on the Fisher discriminant criterion that considers both the between-class and the within-class scatters. By contrast, LTCSP considers maximizing the between-class scatter only. The adaptive extension uses sparse representation to specify the weights between samples in constructing the temporally local scatter matrices. Experiments on single-trial EEG classification confirm the effectiveness of the proposed methods.