What is it about?
Feature extraction is an essential step of spike sorting. Most commonly, it is done using PCA. Here, we study the performance of autoencoders, a type of neural network, as a feature extraction technique within the domain of spike sorting and compare several variants to some of the most used feature extraction techniques.
Featured Image
Photo by Hal Gatewood on Unsplash
Why is it important?
The aim of feature extraction techniques is to create a representation that is unaffected by slight changes in waveform shape as a result of noise. In practice, various phenomena can alter or contaminate the estimated spike shape, such that clusters are not always distinct, but often overlap. It is important to find a technique that is able to extract the most important information that offers separation, while the execution time can be improved by reducing the number of dimensions by ignoring redundant data. The pattern of encoding offered by autoencoders has been proposed to be invariant to noise as they have been demonstrated to be a well-rounded denoising technique providing robustness. Thus, an immediate idea would be to try this approach in the field of spike sorting.
Perspectives
Writing this article was a great pleasure. This article shows that autoencoders can provide better separation that some of the most used feature extraction algorithms within the domain of spike sorting. If you could remember only one thing about this article, remember that: sometimes what everybody uses does not work for you, it is important to keep searching until you find an approach that does.
Eugen-Richard Ardelean
Technical University of Cluj-Napoca
Read the Original
This page is a summary of: A study of autoencoders as a feature extraction technique for spike sorting, PLoS ONE, March 2023, PLOS, DOI: 10.1371/journal.pone.0282810.
You can read the full text:
Contributors
The following have contributed to this page
