Timeline for Should weight decay regularization be divided by the number of samples?
Current License: CC BY-SA 4.0
4 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Feb 18, 2023 at 14:37 | comment | added | Jaume Oliver Lafont | But... in the video the same m divides the sum of losses to get the cost so that is a simpler explanation, to scale the sums accordingly. | |
| Feb 18, 2023 at 14:12 | comment | added | Jaume Oliver Lafont | Maybe the m is assumed to be a constant included in the lambda? | |
| Feb 18, 2023 at 10:52 | comment | added | martinkunev | Thanks, that's a good way to think about it and it makes sense. This still doesn't explain why other sources don't divide by m. | |
| Feb 18, 2023 at 6:42 | history | answered | Jaume Oliver Lafont | CC BY-SA 4.0 |