Can variational autoencoders be used on non-image data?

Question

I have a question about variational autoencoders (VAE),

I need to generate new data from my dataset which contains just numerical data, so i want to use VAE for that task, but all the available tutorials and articles use images as input data for the variatioanl autoencoder.

My question is: can i use VAE for generating new data from my datasets eventhough my data is not images ??

Thank you.

Answer 1

Short answer is yes. You should read up a bit on the basics of neural nets if this wasn't obvious already - an image is just a Channel X Height X Width dimensional vector. You might use different kinds of layers in your network to suit the kind of data that you have to give a better inductive bias, but otherwise nothing changes. Follow those tutorials!