Large Margin Filterin
We address in this paper the problem of multi- channel signal sequence labeling. In particular, we consider the problem where the signals are contaminated by both additive and convolutional noise. We investigate several approaches based on windowing and propose to learn a support vector machine (SVM) classifier and a signal filter jointly. We derive algorithms to solve the optimization problem and discuss different filter regularizers for automated scaling or selection of channels. After considering its properties on a toy dataset, the approach is tested on two challenging real life datasets: BCI time series and 2-dimensional image segmentation. Results show the interest of large margin filtering in terms of performance and interpretability.