A Distributed Online Newton Step Algorithm for Multi-Agent Systems

Most of the current algorithms for solving distributed online optimization problems are based on first-order method, which simple in computation but slow convergence. Newton’s algorithm with fast convergence speed needs to calculate Hessian matrix and its inverse, leading computationally complex. A step is proposed this paper, constructs a positive definite by using information objective function replace inverse method. The proved theoretically regret bound obtained. Finally, numerical experiments used verify feasibility efficiency algorithm. experimental results show that has an efficient performance practical problems, compared several existing gradient descent algorithms.