What are the benefits of using autoencoder to create a filter set over the predefined gabor filter set relative to CNN?

Question

I am working on a convolutional neural network, and I have found various methods of creating filters for convolution of the image. What are the advantages and disadvantages of each?

Answer 1

Using autocoders can in some cases improve performance, provide biologically plausible filters, and more importantly, give you a model based on your data, rather than predefined filters. Autoencoders will give you filters that can best match your data. The only drawback is, of course, the extra computation time. However, in many cases, this does not stop you from studying online.

For additional ratings you can see: http://www.idsia.ch/~masci/papers/2011_icann.pdf