0

Traditionally, Siamese Neural Networks have two inputs. With some tweaking, you can get them to accept any number of inputs. What I don't understand is how to get them to accept variable numbers of inputs. I've seen a couple of research papers (most notably this one) where they talk about doing this, but none explain exactly how.

Could someone please explain how to create a Siamese Neural Network with a variable number of inputs?

GameDungeon
  • 101
  • 4
  • Hello. Welcome to Artificial Intelligence Stack Exchange. Can you please provide the links to the papers that you read so far (that you don't understand)? – nbro Dec 07 '21 at 17:34
  • I've added the main one now, thanks. – GameDungeon Dec 07 '21 at 22:00
  • After more research, I've found that magnet loss could present a solution. I would say more, but I can't find an implementation in keras (the framework I use) and [my question about it](https://stackoverflow.com/questions/70282894/implementation-of-magnet-loss-in-keras) is still unanswered. I'll make sure to edit this if I get a reply. I would love to combine the research done in both the [magnet loss paper](https://arxiv.org/pdf/1511.05939v2.pdf) and the [VIS-CNN paper](https://www.tandfonline.com/doi/pdf/10.1080/13658816.2018.1542698?needAccess=true), but I currently don't have the skill to – GameDungeon Dec 07 '21 at 22:07

0 Answers0