Update loss_surface.md

2026-08-11 11:21:44 +08:00 · 2026-06-11 21:21:01 +08:00
parent 6e9a3ca633
commit 1ad74e14c6
1 changed files with 2 additions and 0 deletions
@@ -68,3 +68,5 @@ When your loss is a product of factors A*B and one factor can be near zero:
 ```

 General principle: if you want gradient to flow independently through two factors, decompose multiplicatively in log space.
+
+You can also design surrogate losses that are better behaved but move in the right direction in a better behaved well.