Hi,
When going through the lecture on gradient descent I noticed that the 3d plot might be a little confusing: it displays values < 0 for J(w,b), when in fact that is not possible. Maybe this can be fixed as it might confuse some students?
It’s an error in the figure. Those are not supported for updates.
Hello ,
Could anyone help me to understand the 3D plot of gradient descent method? As per my understanding, when a person on the top of the figure rotates 360 degree and take a little step, mathematically gradient step, the no of steps he has taken lead to the down hill.. which mathematically infers local minima to minimize J(w, b). But, what is the point of rotating 360 degrees? I mean what it mathematically infer? Also, he has started with one point.. can we start with any other point on the hill? The point which the boy stands, what does mathematically infer? Tell me if I am wrong in my interpretation and sorry for asking this small doubts..3D plot is little bit confusing
I am not sure about this rotation I have not come across it.
But the starting position it has to do with model weights and biases, whatever choice you start with, that can be thought as the starting position in the hill.
Hi @Shivaragavi .
The 360 degree is just an example that Prof. Ng used to illustrate the point- initially, the person standing on top of the hill has no idea which way to start going down, so spinning 360 degree is his way of deciding on a starting point. In other words, the starting point is chosen in random, and so as the weights. Once the starting point is chosen, the gradient decent starts finding its way down the hill. Prof. Ng also stressed that depending on the starting point, the path going down hill is different.
2 Likes
Thank you so much for sharing.
1 Like