Mixture of Autoencoders
The goal of this project is to design a protein classifier inpired by the Mixture of Experts concept. In this case, the experts are autoencoders. The toy example includes a connected stacked autoencoder, a variational autoencoder, and a denoised autoencoder as experts. In addition, there will be a gating function performing soft partitions for each autoencoder. The idea is that if autoencoders became experts in their own domain, then they will perform better classification.