Discussion of article "Neural Networks Made Easy" - page 2

New comment
 
Реter Konow:
The article, part two is "how an artificial neuron works". Correct me if I'm wrong.

1. A neuron consists of several functions:
1. The first one multiplies the input value by a weight coefficient.
2. The second one sums up the obtained products.

Judging from the article - this particular implementation of the neuron contains the following:

1. Weights for all neurons that are included in this neuron.

2. Sums up the weights using the weighted sum.

3. Actualising function - which already outputs the final value of this particular neuron.


Although the code doesn't compile, it is written quite clearly. See the neuron class for details.

 
Okay. Neuron activation function? That is, a function to bring it to a range between 0 and 1, or between -1 and 1?
 
Реter Konow:
Okay. Neuron activation function? That is, the function of bringing to the range between 0 and 1, or between -1 and 1?

Yes, that's right. the author there gave the names of the most frequently used ones in wikipedia or online you can read more about them.

I am also interested in this topic myself, later I will also dig more in detail as time permits).

 
only the input signal must also be in this interval. I mean by input signal - exactly the input signal to the neuron, not to the function under discussion.
 
Andrey Azatskiy:

Yeah, that's right.

OK. It is clear where the input value to the neuron comes from - from the data array or from the previous layer of neurons.
But where does the weight coefficient, by which the input value is multiplied, come from?
 
It seems that the weight coefficients, by which the input values of neurons are multiplied, arise as a result of "training" of the network. That is, they do not exist at first, and then they appear. But how exactly - it is not clear yet.
 
Реter Konow:
It seems that the weight coefficients, by which the input values of neurons are multiplied, arise as a result of "training" of the network. That is, they do not exist at first, and then they appear. But how exactly - it is not clear yet.

I, like you, have not used neurons before, but after reading the article and looking at the code carefully, all such questions have disappeared.
The initial weight value for neurons is set randomly or from a file where it was saved before. Further in the learning process, based on the error of the target value and the value at the output of the most recent neuron, all weights are recalculated. The recalculation of weights itself is performed in each neuron independently (see the part of the article where the neuron is described and view the code of the neuron itself).

 
Andrey Azatskiy:

I, like you, have not used neurons before, but after reading the article and reviewing the code carefully, all such questions disappeared.
The initial weight value for neurons is set randomly or from a file where it was saved before. Further in the learning process, based on the error of the target value and the value at the output of the most recent neuron, all weights are recalculated. The recalculation of the weights itself is done in each neuron independently (look at the part of the article where the neuron is described and view the code of the neuron itself).

I learn material easier from dialogue with knowledgeable people. 10 times faster than from articles.))

Thanks for the clarification.
 
Реter Konow:
The article is interesting. Can you explain "on your fingers":

1. What is a neuron inside?
2. What is a neuron bundle? (Meaning, a connection between two neurons - a layer 1 neuron and a layer 2 neuron).

These are the questions that I did not understand from the article until the end.

Thank you.
Retag Konow:
The article, part two is "how an artificial neuron works". Correct me if I'm wrong.

A neuron consists of several functions:
1. The first one multiplies the input value by a weight coefficient.
2. The second one sums up the obtained products.
3. The third calculates the resulting value in the activation function.

I understand the words, but not the meaning.

Good evening, Peter.
The neuron inside consists of 2 functions:
1. First, we calculate the sum of all incoming signals taking into account their weighting coefficients. That is, we take the value at each input of the neuron and multiply it by the corresponding weighting factor. And add up the values of the obtained products.


Thus, we get a certain value, which is fed to the input of the activation function.

2. The activation function converts the obtained sum into a normalised output signal. This can be either a simple logic function or various sigmoid functions. The latter are more widespread as they have a smoother transition of state change.

Communication between neurons is organised as a direct transfer of the output value of one neuron to the input of a subsequent neuron. In this case, referring to point 1, the input value of a neuron is taken into account in accordance with its weight coefficient.

 
Andrey Azatskiy:
another question what is neurotNam - the method of creating neurons in a layer ? it is not declared anywhere and the logic of why the initial value of a neuron is equal to the remainder of division by 3 minus 1 is not clear ?

The errors in the code have been corrected and the file in the article has been replaced.
The specified line assigns the initial data of the output value of the neuron and can be replaced by a constant. This value will be changed at the first direct recalculation of the neural network value.

123456789
New comment