Principles of Artificial Intelligence- P1 | Sole Concept Lab
Artificial IntelligenceTutorials

Principles of Artificial Intelligence- P1 Concepts needed to get you started with AI

With the growth in the amount of data, super efficient computing methods and also the simplicity in designing and using Neural Networks everyone seems to be doing deep learning!!!!

In these series of blogs, I will walk you through a learning path starting from the principle of Artificial Intelligence to developing ai applications.

Before we begin to understand how AI works, let me walk you through the very very fundamental element of Artificial Intelligence. Neural Network is the very very fundamental unit of Artificial Intelligence. Let’s proceed to understand what exactly are these Neural networks and why the term leading to how ai works.

As with many applications in and developments in science and technology NN’s are also inspired by nature. These are particularly designed to mimic the neurons in the brain. A simplistic and self-explanatory representation of neuron is shown below.

Representation of a Neuron in the brain

Similarly in artificial intelligence also we code a neuron. This neuron performs a specific task or with the incoming data, it is activated to perform a task. I will go through the various commonly used tasks or activation functions below.

Having understood what a neuron is let proceed to understand what a Neural network is. It is pretty self-explanatory isn’t?

You guessed right it’s just a collection of neurons which are connected. To be more specific and heading towards our goal let’s formally define it. A neural network consists of layers of Neurons such that each layer is connected to the other. Let’s take for instance the following network.

Simple Neural Network

Simple Neural Network

The diagram may not be beautiful but it serves the purpose that we are trying to achieve. Firstly each circle, in this case, represents a neuron. The type of network as seen above is called a vanilla network i.e it just has an input layer some processing happens and produces two outputs.

There is also another type of neural network and it can be represented as below (source).

Figure representing multi layer Neural Network

Multi-Layer Neural Network

 

The difference between this and the network previously is very simple. It has one layer between input and an output layer. This layer in between is called hidden layer. There could be one or more layers between input and output i.e there could be multiple hidden layers.

Networks which have layers between the input and output layers are called deep networks and this is widely used in artificial intelligence today.

Having understood the concept of neural network and terminologies let us now start introducing more terms especially weight and bias.

Let’s come back to how a single neuron function with an example. Suppose a neuron receives two inputs a and b both of them are numbers and they range from 1 to 10. The task of the neuron is now to classify the sum of the two inputs as 0 and 1 where 0 stands for the sum less than 5 and 1 for greater than 5. The process is represented as follows.

Example of how a neuron works

Every input is multiplied by a weight. In our case let’s assume the weight to be 0.5 thus input a and b are multiplied by 0.5 and added together. I.e if a and b are 2 we would have the sum as 1+1 =2 this is then inputted to the neuron where a function acts on the input to produce the output (this is called the activation function and will be discussed in detail in the next blog).

But what if the inputs are 0. I understand that they cannot be as I told you that they range from 1 to 10. But you know data isn’t as expected always. If the inputs are zero there is a problem as the sum would be zero. To avoid this error a bias term is introduced. So mathematically the input to the neuron is written as follows

\( y = \sum_{i =1}^2w_i x_i + b\)

In our case here \(w_1 = w_a\) and \(w_2 = w_b\). this input is then fed to the neuron where a function \(\sigma \) acts on y to produce an output. This function \(\sigma(y)\) is what is called an activation function. After processing the information this neuron produces an output classifying the sum as zero and one.

Conclusion

So far we have understood neural networks conceptually and seen an example of how a neuron works and various terminologies. 

Ok, so what are these activation functions and why are they required? Weights how are they important and how is it computed? Confused stay tuned for more.

Sharing is caring!

No Responses

  1. Pingback: Sole Concept Lab May 6, 2018

Leave a Comment