Talk:Capsule neural network

I've finished with this piece now. It is ready for submission and/or publishing. Feedback encouraged. Lfstevens (talk) 02:02, 29 December 2017 (UTC)

I was wondering whether I shoul::::d write this article as I am a bit to opinionated about capsule nets, and why and how it fails compared to cortical minicolumns. Capsule nets as they are now is a bit to much of a hack and to little theory.

In my version I had carefully split out intuition and motivation. As it is now the article seems like the whole idea emerged from nowhere. Those that has followed the field for some time knows that this idea has a history long before the latest papers.

Now the article is about visual representation of objects, which is kind of weird. First of all, spatial relationship between parts of an object is not the object itself. This is about information represented in a vector space, and how transformations on the information can reconnect otherwise disparate points in that vector space. That can be used for reconnecting visual parts of an object, but that is only one of several use cases.

In my version I wrote about how it compares to cortical minicolumns, now it is about a special case where capsule networks are used for interpreting visual scenes. My version was about how capsule nets can be interpreted in the general case, now it is about a special solution for a special problem. It is even worse, as it now presents the specific implementation and its training as the capsule network. The setup is for a specific network where a specific problem is solved, but that is not the general case. And by the way, there is nothing with capsule networks that indicates that it has to use supervised learning. Supervised learning is used in this specific case.

Nearly everything about the inspiration from cortical minicolumns are gone, which is very bad, as it removes the foundation for why capsule networks have some serious flaws. The flaws is how the dynamic routing is done, actually the error is the routing. Inspecting the history it is clear that you simply dismissed that as speculation. Well, to put it bluntly, capsule nets are not how the brain does it, check out what Hinton says about it – he says he doubt this is the way the brain does it, but in my opinion it is a small step in the right direction.

No, I don't think this draft should be posted as it is now. Jeblad (talk) 00:13, 30 December 2017 (UTC)

I don't understand how responding to the points in a post is "messing with it". I presume you read my responses and await yours. Lfstevens (talk) 19:54, 30 December 2017 (UTC)

Do not mess with other users posts. You have already done a rewrite of the text I was working on, changed the meaning, and changed the factual basis. Then you did the same thing with my reply on the discussion page. I guess you believe you have "the right to do it", but I tend doubt you know what you are doing. Jeblad (talk) 02:51, 31 December 2017 (UTC)

While I changed the article a great deal, on this page I merely interspersed my responses to your individual points. That is how I've seen many discussions conducted on WP in the past. I am guilty of removing material from the article that I found to be WP:OR or just unsourced, but that is how WP works. By contrast, the material that I added was all sourced. I may have made errors of fact or interpretation, and if so I am happy to correct them. I have already begun to attempt to incorporate parts of your comments as I find material that supports them. In particular, I am trying to reduce the emphasis on image recognition. Still looking for sources to flesh that out. Please post links, as you like. Lfstevens (talk) 18:10, 31 December 2017 (UTC)

Huh, "changed the article a great deal"? You added factual errors and misunderstandings to the article. Claiming text to be WP:OR, removing it, and then add it back doesn't make me change my opinion! My previous statement was "I tend doubt you know what you are doing" and I still stand by that. Jeblad (talk) 07:32, 2 January 2018 (UTC)

I've let this rest for a month to allow other editors to contribute. Not much has happened. I made another pass over it today, and think it is time to publish. While I'm sure many improvements are possible, I think it's good enough and that publishing is likely to attract the attention of more knowledgeable experts to help get it over the hump. Comments? Lfstevens (talk) 05:54, 12 February 2018 (UTC)

This is unlikely to get improved in Draft where it is hard to find. Legacypac (talk) 10:47, 18 April 2018 (UTC)

Out of date

This article is out of date and die not reflect the current research or scholarship by Hinton. Elephantwalker90 (talk) 04:41, 2 April 2022 (UTC)