11-27-2024, 10:07 AM
(translated by Google translator)
Hello everyone, dear friends.
The discussion of the topic "Threads" gave excellent results.
Now it will be clear to anyone how to use "Threads" correctly.
I dream that we will achieve the same result in the topic "Training the SB-Neuron".
In this topic I don't have a definitive answer to two questions:
Most likely, I am not the only fan of programming in Small Basic who is interested in being able to construct his own neural network by inserting a neuron with special settings into any place in the neural network.
At the moment, we do not have this option. We have access to a neural network consisting of identical neurons. We can only change the number of neurons and layers. This network learns only by changing the weights on the inputs of neurons.
As a result, using such a neural network requires special data preparation. Such a neural network cannot solve many types of tasks. And the accuracy of the responses of such a neural network is not very high.
I invite people who are also interested in this topic to take part in the discussion.
If successful, we will all receive a very interesting tool for our research and entertainment.
Hello everyone, dear friends.
The discussion of the topic "Threads" gave excellent results.
Now it will be clear to anyone how to use "Threads" correctly.
I dream that we will achieve the same result in the topic "Training the SB-Neuron".
In this topic I don't have a definitive answer to two questions:
- Why can we use different Activation Functions, but train a neuron only by changing the input weights? Perhaps the fact that we can use different Activation Functions is a hint that there is a way to train a neuron by changing the parameters of the Activation Function.
- Which learning method is best: changing the input weights; changing the Activation Function parameters; or a combination of these two methods?
Most likely, I am not the only fan of programming in Small Basic who is interested in being able to construct his own neural network by inserting a neuron with special settings into any place in the neural network.
At the moment, we do not have this option. We have access to a neural network consisting of identical neurons. We can only change the number of neurons and layers. This network learns only by changing the weights on the inputs of neurons.
As a result, using such a neural network requires special data preparation. Such a neural network cannot solve many types of tasks. And the accuracy of the responses of such a neural network is not very high.
I invite people who are also interested in this topic to take part in the discussion.
If successful, we will all receive a very interesting tool for our research and entertainment.