Capsule Networks – A group of neurons which uses vectors to represent an object | Hacker News Reader

Capsule Networks – A group of neurons which uses vectors to represent an object

theailearner.com

53 pointsatulkrishna1010 comments

This article is completely wrong. It's already been proven that unless you have the reconstruction penalty, capsules naturally don't capture orientation and work similar to bag of words models.

Do you happen to know where this was demonstrated?

Can you pls give me the links where you have read this(Any research article or blog). Thanks

Please, please do not call activation functions neurons! A neuron is a name for a type of cell in animal brains which has several variants, many modalities of behaviour (chemical, electrical, temporal) and which we are making many new discoveries about as our technology advances.

An activation function is not the only part of an artificial neuron, there's also the weighted input aggregation. Also, it's well known and understood that artificial neurons are biologically inspired, but are not in fact, neurons themselves.

It is amongst people who are interested in such things, but it isn't amongst journalists and many people who read the popular press and get the wrong end of lots of sticks as a result.

That's their fault for not doing their due diligence. And what does it matter to the general public learning of these things through journalists anyway? It's not like they are doing anything with this knowledge that requires precision. It's all intellectual excitement and cocktail party fodder to them.

Does it also bother you when people talk about using a "mouse" to move "files" to their "desktop?"

Only if they use the analogy to do things like convince people that mice carry little red lights on their tummies.

The first sentence:

> Since 2012 with the introduction of AlexNet, convolutional neural networks(CNNs) are being used as sole resource for many wide range image problems.

Um, CNNs have been around for 30 years as neural networks.

Reply on news.ycombinator.com