I'm trying to learn about ANNs. If anyone can tell me my understanding is correct that'd be fantastic.

So if an ANN was to be used for an image classification task - white background with a black alphabetic character written on it. The inputs are each pixel. What I believe is that the ANN learns to increase the weights for the darkened spots and decrease them for the blank background that resides behind the blackened alphabet character. Am I on the right track?

One thing is certain... learning occurs by the changing or re-assigning of weights.

How to train the network is another field with many options... self organising maps or user based training.

The method I mentioned - is that a valid option from the many options you mentioned?

Without going into details I would say that is a valid option. The one you described.

If you get a chance go through youtube jeff heaton's channel. He is a champion in AI and translating it to code dotnet and I believe java.

I'll take a look. Thank you so much!