Feature Learning for Binary Data

I'm trying to do feature learning / dimensionality reduction on a large, sparse binary dataset. I tried RBMs and then Autoencoders but was dissapointed to find their behaviors roughly equivalent. I thought, since AEs are not restricted to a binary hidden representation, they might have more representational power. Not the case.

I started looking at Logistic PCA and latent, probabilistic models like this one. Does anyone have pointers to other techniques? I assume entropy-based techniques could also be useful?