Timeline for Why is step function not used in activation functions in machine learning?
Current License: CC BY-SA 3.0
12 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| S Aug 7 at 4:49 | history | suggested | Super Kai - Kazuya Ito |
I added two tags.
|
|
| Aug 6 at 18:44 | review | Suggested edits | |||
| S Aug 7 at 4:49 | |||||
| Aug 6 at 18:30 | answer | added | Super Kai - Kazuya Ito | timeline score: -1 | |
| Jan 1, 2022 at 16:28 | history | protected | kjetil b halvorsen♦ | ||
| Jan 1, 2022 at 16:19 | answer | added | PING | timeline score: 0 | |
| Sep 10, 2020 at 17:10 | answer | added | Sycorax♦ | timeline score: 7 | |
| Sep 29, 2018 at 1:32 | vote | accept | curious | ||
| Dec 14, 2017 at 11:11 | answer | added | andfor | timeline score: 13 | |
| Dec 14, 2017 at 5:21 | answer | added | bohr | timeline score: 18 | |
| Apr 4, 2017 at 8:16 | comment | added | itdxer | @AlexR. I think it's a pretty good and simple answer and you should right it as an answer instead of the comment. | |
| Apr 4, 2017 at 1:23 | comment | added | Alex R. | Assuming you're talking about the heaviside step function, it has 0 gradient everywhere except 0, so you could never do backpropagation since your gradient would always be 0. | |
| Apr 4, 2017 at 0:39 | history | asked | curious | CC BY-SA 3.0 |