Exploiting negative curvature directions in linesearch methods for unconstrained optimization

摘要: In this paper we propose efficient new linesearch algorithms for solving large scale unconstrained optimization problems which exploit any local nonconvexity of the objective function. Current in class typically compute a pair search directions at every iteration: Newton-type direction, ensures both global and fast asymptotic convergence, negative curvature enables iterates to escape from region non-convexity. A point is generated by performing along line or curve obtained combining these two directions. However, almost all if algorithms, relative scaling not taken into account. We algorithm accounts To do this, only most promising selected given iteration, performed chosen direction. The appropriate direction estimating rate decreas...