Creation of SB-Neuron. Ours. Branded.(v2)

[z-s]

Good job my exams are over now

[AbsoluteBeginner]

Good job my exams are over now

Congratulations to you! 

What are you planning to create now? 

[litdev]

When trying to draw with the mouse button pressed, there are problems with synchronization. 

What requirements must be met when organizing the simultaneous operation of an event handler and the process of drawing an image on the Graphics window? 

All interface updates only occur on the main UI thread, so best to not do this in an event subroutine as they only get done when the event subroutine ends and this can cause issues in an event that occurs often or takes some time.

The recommended approach is to set flags in event subroutines and do the UI work in a message loop.

This is an article I wrote on this for the old forum, https://learn.microsoft.com/en-us/archiv...ent-basics.

[z-s]

Good job my exams are over now

Congratulations to you!  

What are you planning to create now? 

Some finishing in ZS extension then a bigger project 

[AbsoluteBeginner]

I have updated the code of the "NeuroLab" program. 

Now you can change the color of pixels with one click or by moving the cursor while holding down the mouse button.

DWXL365.000-0

Next, together with Gemini, we will continue to develop this interesting project. 

We wish you all the best of luck.

[AbsoluteBeginner]

Hello. 

Four days ago, my chat with Gemini, which was set up to work with the project, started producing all sorts of nonsense.
For two days I tried to fight this.
For two more days I waited for the problem to disappear on its own.

Today I created a new chat to continue working on the project.
The first results show that the chat is working normally.
Therefore, I hope to publish the next version of the code soon. 

Good luck to you all.

[AbsoluteBeginner]

My hopes were dashed. 

Weak Gemini SB-programmer.

He's even more of a "Beginner" than I am. 

[AbsoluteBeginner]

Hi all. 

Until yesterday evening, there was only one type of SB-neurons in my consciousness.
This was the type in which fixed weights of the neuron's inputs created a precise x-coordinate scale for the Activation Function graph.
The training of such a neuron consists of simply writing the training data into the Activation Function about the coordinates X, which are specified by the data at the neuron's inputs.
Such neurons are good at determining the nature of the DEPENDENCE of output values on input data.

However, there is another class of problems in which it is necessary to RECOGNIZE an object, or a current situation, based on an input SET OF FEATURES.
In this case, the input data of the SB-neuron should not create the exact X-coordinate from which the Activation Function will determine the output value.
In this case, the input weights must indicate the IMPORTANCE of each feature so that the neuron can confidently determine that the current situation corresponds to the situation that the neuron was trained on.

Such a SB-neuron already closely resembles the classical neurons that are used in the ANN extension.
I'll call it "SB-neuron Type-B". 
But the internal structure of such a SB-neuron will differ from the classical one.

[AbsoluteBeginner]

Hi all. 

When I develop my NeuroLab program, I try to find ways to reduce the number of SB-neurons that are required for the program to function properly.
I was looking for a way that can find FEATURES in the information contained in an image on a 10x10 pixel matrix.

And today, an interesting idea came into my head.

Imagine this picture:
What happens if you expand the pixel matrix into one row? Then, "roll" this line into a ring, connecting the beginning of the line with its end.
Let this be the TEMPLATE of the figure "Cross", which we saved in the memory of the "SB-neural network".
Now, if we draw the same “Cross” on the “SB-display” of the “NeuroLab” program, but located in a different place of the matrix, then we can also “roll up” the data row of this matrix into a ring.

Let's now place both rings next to each other so that the "Template" is positioned ABOVE the new image.
Let first the pixel numbers of the “Template” and the new image coincide vertically. That is, the first pixel of the "Template" is located ABOVE the first pixel of the new image.
Now, we will count how many times the color of the "Template" pixel matches the color of the pixel of the new image located directly below it and write down this result.

After this, we shift (rotate) the lower ring with data by one pixel so that the second pixel of the new image is located under the first pixel of the "Template".
We will count the number of matches again and write down the result again.
We repeat this operation until the lower ring makes a "full turn" and the first pixel of the "Template" is ABOVE the last pixel of the new image.

Now, if we look at the records we made, we will surely see that the MAXIMUM number of matches between the color of the "Template" pixels and the pixels of the new image will be when the data of the "lower ring" coincides in space with the data of the "upper ring" (Template).

But the most interesting thing for me is to see how the "coincidence graph" will look if the new image also contains a "Cross", but whose size DIFFERENT from the size of the "Cross" on the "Template".
Will the "FEATURES" of the "Cross" figure be able to manifest themselves in the form of local maxima on the "Chart of the number of coincidences" when the size and position of the two crosses do not match?

[AbsoluteBeginner]

Hi all. 

Today I launched and tested the next version of my NeuroLab program.
   
In the screenshot, it shows a graph of the values of the weighted sums at the neuron inputs when comparing images of a small and large circle.
At this point, when the inputs of all neurons have a weight of 1, many of the neurons have the same weighted sum, despite the fact that the location of the large circle relative to the small one differs for each neuron.
The next step in developing the program will be to create the ability to set the value of the weights on the inputs of neurons to see how this changes the shape of the graph.
The user will be able to learn how to change the input weights so that ONLY ONE largest local maximum is formed on the graph.