Talk:Siamese neural network

Euclidean distance (Manhattan distance)

Euclidean distance (Manhattan distance) is not the same as squared Euclidean distance (squared Manhattan distance), the former is a first order linear distance metric while the later is a second order non-linear distance metric. It is although monotonic and continuous above zero.

Seems like two wrongs have given one right in this case. ;) Still note that a second order non-linearity can give rise to instability during training, the system will drift into instability, but that is an other problem. Jeblad (talk) 12:26, 15 October 2019 (UTC)[reply]

Age of the term

The term is pretty old, and Google books has at least a reference to lecture notes back in 1996. I'm pretty sure I heard about it back in 1992-93, and back then it was still old. I can't remember how it was defined though, and how it was trained. The book we used at UiO (ISBN 0-201-51560-151560) does not mention the network type. (Could be some seminar?) What is pretty clear though, is that my references to papers from 2004 and 2014 is not the first references at all, and that the reference to a paper in 1994 is probably neither the first one. Jeblad (talk) 12:59, 15 October 2019 (UTC)[reply]

Siamese network

I usually hear references to this as “Siamese network”, not “Siamese neural network”. Later name is probably better to avoid name clashes, but still it should probably be mentioned. Jeblad (talk) 13:44, 12 November 2019 (UTC)[reply]