74 lines
1.6 KiB
Markdown
74 lines
1.6 KiB
Markdown
# Concepts of a Neural Network (NN)
|
|
|
|
simulation of human nervous system in its ability to learn and adpt.
|
|
|
|
## Units
|
|
|
|
1. Inputs
|
|
2. Weights
|
|
3. Transfer Function + net output
|
|
4. Activation Function
|
|
5. Outputs
|
|
|
|
### Flow
|
|
|
|
1. Update and Provide our Inputs;
|
|
2. Transfer Function;
|
|
3. Net OutPut (Inputs and Weights);
|
|
4. Actiation Function;
|
|
5. Calculate Output;
|
|
|
|
### Learnign
|
|
|
|
it is a process to Change Weights of inputs based on results ...
|
|
this happens to algorithm improve itself optimization ...
|
|
|
|
### Net
|
|
|
|
in this process Inputs and their Weights collecting together.
|
|
|
|
### Activation Function
|
|
|
|
Recieved Net Inputs (Weighted Inputs) and then Calculate output based on them.
|
|
|
|
## Multi Layer NN
|
|
|
|
1. Input Layer;
|
|
2. Hidden Layer;
|
|
3. Output Layer;
|
|
|
|
Hidden Layers recieved Data from all other layers at the end, then Populate as a Neuron.
|
|
|
|
### Input Data Normalization
|
|
|
|
a Process where all the input data normalized.
|
|
reduce data to an accepted ranges.
|
|
i.e. [0,1] or [-1, 1].
|
|
|
|
this peocess so important for us for making data more acceptable.
|
|
|
|
this can done by some standard form:
|
|
|
|
y = x - (x(min) * (d2 -d1)) / x(max - x(min)) + d1
|
|
|
|
x => value to normalized;
|
|
x(min)/x(max) => x Value range max and min;
|
|
d1,d2 => ranges to upper and lower normalization;
|
|
|
|
### Activation Functions
|
|
|
|
a function which calculate the output of a neuron.
|
|
it recieved a Net Input (Weighted functions);
|
|
|
|
1. Unit Step / Hard Threshold Functions;
|
|
2. Sigmoid Function;
|
|
3. Hyperbolic Tangent Function;
|
|
|
|
## Unti Step
|
|
|
|
a Graph by y axix Output and x axix Net Input and values in y between 0, 1 and x is 0 to TETA or threshold.
|
|
|
|
### Sigmoid
|
|
|
|
Change Activation Functions Shape;
|