Leaky Relu solves the problem of dead neurons. Because it is not zero even in the case of negative values. Let’s see leaky Relu derivative python.
Let’s see the mathematical expression for Leaky Relu.
x>0 then f(x)=x
x<0 then f(x)=x*constant
Here we have seen that the value of Leky Relu function is always the same as the variable if the value is greater than zero. It will be a constant in the place of zero. Now let’s see the mathematical expression of the derivative of Leky Relu.
x>0 then f(x)=1
x<0 then f(x)=constant
This constant is the only difference between Leaky Relu and Relu. In the below section, we will see its importance.
In the above section, We have seen the mathematical expression. Now let’s see leaky Relu derivative python Implementation
def leaky_Relu(x):
return x*0.01 if x < 0 else x
def leaky_Relu_Derivative(x):
return 0.01 if x < 0 else 1
In the earlier section, We have seen the mathematical expression and python code for Leaky Relu. Now let’s see the graph of Leaky Relu.
Let’s see the Graphical Representation of Leaky Relu Derivative. Here we need to be careful that it looks approximate to zero in negative values. But actually, it is not zero.
Leaky Relu is a Revolution in Neural Network. It solves the problem of Vanishing Gradient Descent in RNNs. That is a clear reason for rising in the Deep Learning journey. Actually, Sigmoid Function’s derivative has a range between (0,0.25) which tends to zero because of the chain rule. This causes the deactivation of neurons.
I hope now must have understood an intuition of Leaky Relu. How its existence helps neural networks. Still, If you have any doubt on Leaky Relu, Please comment below in the comment box. We also appreciate you for reading the whole article till the end.
Thanks
Data Science Learner Team
Reference
1 .https://paperswithcode.com/method/leaky-relu
2. https://www.analyticsvidhya.com/blog/2020/01/fundamentals-deep-learning-activation-functions-when-to-use-them/