Timeline for Asymptotic behavior of gradient descent on a smooth, convex, non-negative function with no finite minimum - part II
Current License: CC BY-SA 3.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Nov 21, 2022 at 2:52 | comment | added | Jean Legall | I think you may need to require $f$ to be a $C^1$ semi-algebraic function. | |
| Oct 9, 2017 at 6:30 | history | edited | Daniel Soudry | CC BY-SA 3.0 |
added 34 characters in body
|
| Oct 7, 2017 at 19:42 | comment | added | Daniel Soudry | Yes, 'Hessian' would be clearer here, corrected. I don't think it is this easy though. The Hessian can vanish to zero at infinity, yet the ratio can remain bounded. | |
| Oct 7, 2017 at 19:36 | history | edited | Daniel Soudry | CC BY-SA 3.0 |
added 6 characters in body
|
| Oct 7, 2017 at 19:34 | comment | added | Surb | You mean the "hessian" should have a bounded eigenvalue ratio, right? This is definitely a sufficient condition as it implies that the function is then strongly convex and thus have a unique global minimizer. So your gradient descend will converge to this solution. | |
| Oct 7, 2017 at 15:06 | history | asked | Daniel Soudry | CC BY-SA 3.0 |