Timeline for Training vs test accuracy trade-off
Current License: CC BY-SA 3.0
3 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 4, 2018 at 19:16 | comment | added | Sixiang.Hu | Get rid of noise fully is not really practically imo. The reason is "get rid of noise" fully means you can "model" noise ( so that it can be regularized ), which is not practical. And this is why I used "control fitting to noises" as oppose to "fully get rid of noise". | |
| Apr 4, 2018 at 19:10 | comment | added | sdiabr | So @Sixiang.Hu, assuming that point 2. does not exist in this case, i.e. training and test data are rather simmilar, would then be always possible to get rid of that noise/overfitting gap by applying the correct amount of regularization? | |
| Apr 3, 2018 at 9:49 | history | answered | Sixiang.Hu | CC BY-SA 3.0 |