%0 Journal Article
%T LOSSGRAD: Automatic Learning Rate in Gradient Descent
%A Wójcik, Bartosz
%A Maziarka, Łukasz
%A Tabor, Jacek
%J Schedae Informaticae
%V 2018
%R 10.4467/20838476SI.18.004.10409
%N Volume 27
%P 47-57
%K gradient descent, optimization methods, adaptive step size, dynamic learning rate, neural networks
%@ 1732-3916
%D 2018
%U https://ejournals.eu/en/journal/schedae-informaticae/article/lossgrad-automatic-learning-rate-in-gradient-descent
%X 
	In this paper, we propose a simple, fast and easy to implement algorithm LOSSGRAD (locally optimal step-size in gradient descent), which automatically modifies the step-size in gradient descent during neural networks training. Given a function f, a point x, and the gradient ▽xf of f, we aim to find the step-size h which is (locally) optimal, i.e. satisfies:

	h = arg min f(x - t▽xf).
	            t≥0

	 

	Making use of quadratic approximation, we show that the algorithm satisfies the above assumption. We experimentally show that our method is insensitive to the choice of initial learning rate while achieving results comparable to other methods.