On Earth, Human brain is the most powerful learner and neurons are the ones which give them this capability.Below diagram shows the basic structure of single neuron present in human brain.Here dendrites are the receivers who receive electrical signals from other neurons and axon is the transmitter of the neuron which is connected to the dendrites of other neuron and likewise.
One individual neuron isn’t powerful but when millions of them are interconnected and work together they can do wonders.
Coming to the technology side,The whole concept of deep learning and neural network works on the same ground and try to recreate this behavior of human brain.
Neuron is the basic building block of artificial neural network which have input values similar to the dendrites and an output value similar to axon.Neuron is typically represented by a circle with some input and output values.This input values x1,x2……xn are the features or independent variables and output is the label or dependent variable.
For example if you want to predict the resale price of a car, then car model,make,manufacture,airbags,engine cc etc.. are the features or input and price is the label or output.
Every single neuron in neural network has two basic functions.One is to compute linear function and other is to perform non linear function.To make it simple we can say one part of the neuron compute linear function and after that the second part will compute the non linear function.
Now another crucial entity in neural network are weights.Every input to the neuron has weight associated with it.Training a model is all about adjusting these weights and its crucial because value of these weights decides how your neural network is going to learn or how accurate your model is going to be.
Computation of Linear Function
In first step every input value gets multiply with its associated weight and are added up to give weighted sum and perform linear calculation.Its linear calculation because the weights (W1,W2…Wn) are linear in nature with power of 1.
Z = x1*w1 + x2*w2 +x3*w3 +……………xn*wn
Or it can be represented as,
Computation of Non Linear function
Once the linear computation is completed,In the next step non linear computation will be performed using activation function.There are couple of activation function available,If i have to name some of them they are sigmoid,Hyperbolic tangent(tanh),rectifier etc…
Let’s move ahead with sigmoid function in this case which is typically a choice for binary classification problem.Its a ‘S’ shaped curve and the output value lies between 0 and 1.
Where Z = x1*w1 + x2*w2 +x3*w3 +……………xn*wn
In the next article we will understand this in depth and will include more entities of neural network.
Thanks so much for sparing your time and reading.This is my first article on medium and will bring up more on machine learning,deep learning and artificial intelligence.Let me know in a comment if you felt like this did or didn’t help.
Clap it! Share it! Follow me!