Bounded Projection Matrix Approximation with Applications to Community Detection

Community detection is an important problem in unsupervised learning. This paper proposes to solve a projection matrix approximation with additional entrywise bounded constraint. Algorithmically, we introduce new differentiable convex penalty and derive alternating direction method of multipliers (ADMM) algorithm. Theoretically, establish the convergence properties proposed Numerical experiments demonstrate superiority our algorithm over its competitors, such as semi-definite relaxation spectral clustering.