Types of Activation Function
The activation function is the most major factor in a neural network which chose whether or not a neuron will be actuated or not and assigned to the coming level. This exclusively means that it'll choose whether the neuron’s input to the network is applicable or not in the procedure of forecasting. For this explanation, it's also appertained to as threshold or metamorphosis for the neurons which can meet the network.
The activation function can be extensively classified into 2 classifications.
- Binary Step Function
- Linear Activation Function
A binary step function is commonly applied in the Perceptron linear classifier. It thresholds the input values to 1 and 0, if they're higher or lesser than zero, apart.
The step function is basically applied in binary classification challenges and works well for linearly severable pr. It ca n’t classify the multi-class problems.
The equation for Linear activation function is
f (x) = a.x
Properties
Range is- infinity to infinity
Provides a convex error face so optimisation can be attained quickly
df (x)/ dx = a which is stable. So can not be optimised with grade descent
Limitations
Since the derivation is constant, the grade has no relation with input
Back propagation is constant as the change is delta x
Non-linear activation functions
current neural network models apply non-linear activation functions. They allow the model to produce complicated mappings between the network’s inputs and outputs, similar as images, video, audio, and data sets that are non-linear or have high dimensionality.
Majorly there are 3 types of Non-Linear Activation functions.
Sigmoid Activation Functions
Sigmoid Activation is a function which is compassed as ‘S’ shaped graph.
Equation
A = 1/ (1e-x)
Rectified Linear Units or ReLU
Learn what is relu here.
Complex Nonlinear Activation Functions
Conclusion
Here, we learned about activation function, its properties ,limitation and types of activation function.
Comments
Post a Comment