relevant sparse code
Reviews: Relevant sparse codes with variational information bottleneck
I find the paper novel and interesting. To my knowledge the algorithm is original and it adds to the existing tollbox of IB based approaches. The proposed method seems to outperform Gaussian IB on denoising and occlusion/inpaiting tasks on simulated and real data. It also provides new analysis tools for sparse representations in the form of IB information curves. Overall I think this work has many promising applications in machine learning and neuroscience and would be of interest to the NIPS audience.