How to overcome plateaus in gradient descent?

rexbenjamin

New member
Joined
Oct 9, 2008
Messages
2
Reaction score
0
Points
1
Let's say for example, I am utilizing gradient descent to find the minimum of x^10.

The gradient is 10x^9. With an initial guess of 1, I can get to 0.1. The true minimum is 0, but the gradient at 0.1 is 10*(0.1)^9 which the computer thinks is zero.

How can i get over this plateau?
 
Back
Top