8a3363abe792db2d8761d6403605aeb7-Reviews.html

This paper proposes an algorithm for normalized cuts of hypergraphs by formulating the cut as the minimization of a ratio of two convex functions which can be solved using existing methods (RatioDCA, with an inner problem solved using a primal-dual method). Semi-supervised learning on a hypergraph is formulated as a related optimization problem and solved with a similar primal-dual method. The proposed approach is shown on several datasets to outperform an alternative technique based on a transformation of the hypergraph to a regular graph for a semi-supervised learning, a clustering and a cut objective. The paper is clear and well written. It is technically sound and provides a significant contribution to the problem of hypergraph cut, and possibly to semi-supervised learning and clustering --- assuming a hypergraph based approach is relevant to the problem. Concerning this last point, not much is said about the relevance of the hypergraph approach.