AN ACCELERATED METHOD OF GRADIENT DESCENT

135 views

Abstract

Optimization problem is the problem of finding the best solution in best solutions. Optimization has many applications in deep learning real life such as classification problems, image recognition, problems to maximize revenue or reduce costs, production time vv. The gradient descent method usually used to find the optimal solution of a problem quickly. In this report, the author uses a new method to quickly find reasonable learning rate based on the idea of ​​flow control and anti-congestion principle in telecommunication networks to speed up the ability convergence of the problem compared to the conventional gradient descent method.