Timeline for What does it mean that classes are mutually exlcusive but soft-labels are accepeted?
Current License: CC BY-SA 4.0
8 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| May 25, 2018 at 14:44 | history | rollback | Green Falcon |
Rollback to Revision 2
|
|
| May 23, 2018 at 14:22 | comment | added | ignatius | In this link there is a nice discussion about this topic, and it covers the rare case in which for multi-class classification, soft-labels are accepted <stackoverflow.com/questions/47034888/…> | |
| May 23, 2018 at 12:34 | comment | added | ignatius | Thx, I was thinking in something similar. My conclusion is: although softmax_cross_entropy_with_logits needs labels to be a valid probability distribution (not-only the one-hot encoding) it is very rare that for a mutli-class case the probability distribution of the labels are not of this form. So, as a rule of thumb, use this loss-function with one-hot encoded labels (though not necessary, but rare). | |
| May 23, 2018 at 12:19 | history | edited | Green Falcon | CC BY-SA 4.0 |
edited tags
|
| May 23, 2018 at 12:19 | answer | added | Green Falcon | timeline score: -2 | |
| May 23, 2018 at 11:42 | comment | added | Fadi Bakoura | There are two distributions (ground truth, model distribution), we use the cross-entropy loss function to measure their distance. thus for the backpropagation to work, the probability distribution of the labels doesn't have to be one hot encoded. for the exclusive case maybe they are modeling the noise (difficult examples). | |
| May 23, 2018 at 10:54 | history | edited | ignatius | CC BY-SA 4.0 |
added 379 characters in body
|
| May 22, 2018 at 14:50 | history | asked | ignatius | CC BY-SA 4.0 |