Artificial Neural Networks in Java

[asif49]

Hi,

I'm really interested in things like image recognition and certain other aspects that Artificial Neural Networks perform really well at and I'm hoping to do a personal project with ANNs in the future. The problem however is that my knowledge of them isn't that elaborate at the moment. All the tutorials I have found to date explain with extremely complex equations, and equations aren't usually a problem for me if they are explained well, but some of these books/online tutorials seem to have 60% equations to 40% text, where the text already assumes a great deal of understanding of ANNs and complex mathematics in general.

I'd like to start at the bottom with first building a single layered network in Java as I think that will be simpler and will provide me with a good standing to approach the multi-layered one from that point onwards but I need good sources to learn from, ones which doesn't make huge assumptions about their readers.

Also, I understand that there are some frameworks that can be used with Java such as Joone but this will not help my understanding of how neural networks can do what they do and for that reason I'd like to develop one from scratch. Can anyone point me to some helpful links or tell me how to go about learning the whole process of ANN development/ what approach is best for image recognitions/which algorithms to use/ ANYTHING!

[Taywin]

The AI field involves a lot of mathematics & statistics (more on statistics though). It is very difficult to explain if you don't have knowledge base of those requirements. If you don't get it at all, you may not be ready to deal with this subject yet. If you insist on learning about it, you may need to go very slow on it. If you see an equation you don't know, stop and look for the meaning of equation before you move on.

In this topic, each node has its probability value calculated from an equation (learning). Normally, you may not need to know where the equation is from, but you would want to understand why it is set that way. You could also adjust alpha value in order to fix your node learning ability. Then you will have to choose a learning algorithm. Once you get both running, you would have somewhat enough resources to start an application of you want to do.

PS: Don't be too rush into this field. Also, you may get discouraged because you may feel disconnected while working in this field. The field is not for everyone and not everyone who is interested in the field could understand it...

[asif49]

It doesn't have to be that mathematically complex as it is made to seem. I think these lecturers need to value the sharing of knowledge among people of all fields and not just the mathematically advanced - and should word their papers/books with this in mind.

Having said that I've been reading several books/articles/youtube videos and am quite sure I'm now able to go ahead an implement a single layered feedforward supervised learning neural network that learns using Hebbian learning. One of the videos that was of great help was this; http://www.youtube.com/watch?v=6ixqKw7uK6o

Those who feel in the same situation should watch that video, and read Phil Picton's book on neural networks which although outdated manages to explain some complex topics with ease.