For people who are confused what is going on: as he's not providing much description of the theory here, it's understandable that one might be confused. Essentially, he has set up a problem to solve that consists of two things: images of handwritten characters (which in this case are features or feature vectors of the problem, to use standard terminology) and the actual, previously known value of those characters 0-9A-Za-z (which are generally called labels). These are fed as the input and expected output into a neural network. The goal is to train the network to learn how to go from image -> character value (from "feature" -> "label") based off of this data.

Very intresting program.

I can't find the OutputColumnHint in my output layer

Awesome!

Skynet next! just kidding, this seems really cool.

Good tutorial, but you got to define what you do and why or the viewer will not really learn anything.

I got this working on my home machine and it seems pretty powerful.  Still have to play around with it more and try creating my own datasets.

So, I really didn't get what he was doing here but would you be able to make something with this that would be able to solve say a puzzle or get out of a maze with this? Assuming you had the skill to set everything up for this? I wouldn't do anything like that because I have no idea what is going on but I am just curious as to what this allows you to do.

You sound very very Danish, also it would seem that this tutorial is not really compatible with the newest version, some settings have clearly changed, and the observers aren't displaying the correct values.

i can't read anything because the quality is so bad

ok this is above me

this is basically a visual scripting tool

I can't find the OutputColumnHint in my output layer