1: \begin{abstract}

2: The growing demand on efficient and distributed optimization algorithms for large-scale data stimulates the popularity of Alternating Direction Method of Multipliers  (ADMM) in numerous areas. While linear equality constrained problems have been extensively explored to be solved by ADMM, there lacks a generic framework for ADMM to solve problems with nonlinear equality constraints, which are common in practical applications (e.g., orthogonality constraints). To address this problem, we proposed a new generic ADMM framework for handling nonlinear equality constraints in this paper, called neADMM. First, we propose the generalized problem formulation and systematically provide sufficient conditions to ensure that each subproblem of the neADMM has an optimal value. Second, we give an upper bound of the objective value and prove the convergence properties under a special condition. In addition, two useful nonconvex applications are shown to fit into our neADMM framework. Extensive experiments on several datasets demonstrate the convergence as well as the effectiveness of our neADMM against several state-of-the-art methods.

3: \end{abstract}

4: