PCA works well sometimes but doesn't actually optimize the problem that we'd like to solve - separating similar from dissimilar features. Some other methods might work better for their separation and are relatively easy to implement. A CNN could also work with a final layer projecting into two dimensions and might work reasonably well?
PCA works well sometimes but doesn't actually optimize the problem that we'd like to solve - separating similar from dissimilar features. Some other methods might work better for their separation and are relatively easy to implement. A CNN could also work with a final layer projecting into two dimensions and might work reasonably well?