Timeline for Why does the sign of the loss in VAEs appear to be backwards?
Current License: CC BY-SA 4.0
7 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jun 10, 2023 at 22:01 | history | edited | Sycorax♦ | CC BY-SA 4.0 |
added 44 characters in body
|
| Jun 10, 2023 at 21:51 | history | edited | Sycorax♦ | CC BY-SA 4.0 |
qu
|
| Jun 5, 2023 at 12:45 | comment | added | Sycorax♦ | In equation 3, which is a likelihood, and therefore is maximized, KLD has a negative sign. Or, if you like, write down equation (3) with reversed sign and recall that it is a loss, so it is minimized. | |
| Jun 5, 2023 at 9:39 | comment | added | SvenG | I am still struggle to understand get an intuitive understanding of the two loss components. I get that we want to maximize the log-likelihood term (reconstruction loss), but why are we maximizing the KL-term (regularization loss)? Do we not want the KL to be as close to 0 as possible, i.e. minimize it? How can we maximize both a maximization and a minimization term at the same time? or does "maximize the KL term" mean "maximizing a minimization problem means we are maximizing it, if we the term is close to 0"? | |
| Jun 4, 2023 at 20:38 | history | edited | Sycorax♦ | CC BY-SA 4.0 |
added 2 characters in body
|
| Jun 4, 2023 at 19:57 | history | edited | Sycorax♦ | CC BY-SA 4.0 |
added 1086 characters in body
|
| Jun 4, 2023 at 19:46 | history | answered | Sycorax♦ | CC BY-SA 4.0 |