Reviews: The Sparse Manifold Transform

This paper presents an unsupervised learning method called sparse manifold transform for learning low-dimensional structures from data. The proposed approach is based on combining sparse coding and manifold embedding. The properties of the learned representations are demonstrated on synthetic and real data. Overall, this paper is a long-winded presentation of a new manifold learning approach, which (at least in my view) is explained in heuristic and vague terms and seems very hard to perceive. Conventional manifold learning makes the assumption that the high dimensional data points lie close to a low dimensional manifold, and different manifold learning methods find low-dimensional embeddings that preserve different local/global properties of data.