Sparse Unsupervised Capsules

I thought I’d share recent work with fellow students who might be interested in unsupervised learning, and have been following recent research on capsules [1, 2]. If you haven’t heard of capsules before, you can watch this video about it, this post series by @maxim.pechyonkin or other resources mentioned in the previous thread on supervised capsules.

My colleagues and I at Project AGI put up this paper a few days ago demonstrating our work to create capsules network based on [1] that is trained in an unsupervised manner and retain the desirable qualities from its supervised counterpart.

I would appreciate any feedback from you guys, and happy to answer any questions! :slight_smile:

Blog post:
Reddit thread: