Early stopping ONLY depends on the 15% validation subset performance. NOT on the 70% training or 15% testing performance.
Increasing val set error is proof that the net is not generalizing well to nontraining data.
Since the decreasing improvement neither refers to the 70% training or 15% testing subset, there is no proof that there is a need to back up.
Hope this helps.
Thank you for formally accepting my answer
Greg
0 Comments
Sign in to comment.