Timeline for Training a Siamese Neural Network for object similarity assessment
Current License: CC BY-SA 4.0
10 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Dec 19, 2019 at 21:13 | comment | added | Daniel | What is your reasoning behind such bumps during the learning process, though? | |
| Dec 19, 2019 at 16:00 | comment | added | Daniel | I will look more thoroughly into the covariate shift aspect, since I think the diversity of objects might be too big for the machine to learn. I believe early stopping could not really be applied in this case. | |
| Dec 19, 2019 at 15:59 | vote | accept | Daniel | ||
| Dec 19, 2019 at 13:26 | comment | added | Noah Weber | Yes, to a degree of course. You cant predict patterns that you did not learn on | |
| Dec 19, 2019 at 13:26 | comment | added | Daniel | When you mention covariate shift, do you mean that similar and dissimilar pairs should be 'similar' amongst the three sets? | |
| Dec 19, 2019 at 13:09 | history | edited | Noah Weber | CC BY-SA 4.0 |
added 579 characters in body
|
| Dec 19, 2019 at 13:00 | comment | added | Daniel | No, so the training set contains approx 50,000 dissimilar, 50,000 similar pairs | |
| Dec 19, 2019 at 12:59 | comment | added | Noah Weber | Yes, thats important info. I deduced from the former that you only have dissimiliar objects==negative pairs | |
| Dec 19, 2019 at 12:58 | comment | added | Daniel | Thanks for your comment. I am not sure what you mean by negative pairs, but the training set I used is balanced - rephrased the explanation. So it should be as exposed to similar as to dissimilar objects. | |
| Dec 19, 2019 at 12:55 | history | answered | Noah Weber | CC BY-SA 4.0 |