Application of feature subset selection on meaningful features leads to promising results for spike sorting
Details
- Supervisors
- Faculty
- Degree label
- Abstract
- Understanding the visual system is a real interest since it contributes to one of the most important sense for human: the vision. To achieve this goal and with a focus on neuronal responses, spike sorting is generally applied on electrophysiological recordings of neural activity evoked by light stimuli. In this process, Principal Component Analysis (PCA) represents the tool commonly utilized for the feature extraction step. However, results are not perfect and knowledge on the phenomenon is not taken into account in this technique. Based on that, this work proposes a set of features directly extracted from the spike waveforms on which feature subset selection algorithms are applied. It shows moreover that these embedded methods, using the best of filter and wrapper tools, lead at least to similar results than PCA, and sometimes to better performance levels. Furthermore, a new technique for the selection of relevant features is proposed. This combination could give rise to a decrease of the time required to apply spike sorting due to the dimension reduction it allows...