Code360 powered by Coding Ninjas X Naukri.com. Code360 powered by Coding Ninjas X Naukri.com
Table of contents
1.
Introduction
2.
ANN and how it works
2.1.
The input layer
2.2.
The hidden layer
2.3.
The output layer
3.
Some major processes in ANN
3.1.
Forward Propagation
3.2.
Backpropagation
3.3.
The activation function
4.
Advantages and Disadvantages of ANN
5.
Applications of ANN
6.
Frequently Asked Questions
7.
Key Takeaways
Last Updated: Mar 27, 2024

Artificial Neural Networks

Author Arun Nawani
0 upvote
Master Python: Predicting weather forecasts
Speaker
Ashwin Goyal
Product Manager @

Introduction

AI, as we know, is one the most in-demand technology today and might stay the same over several more years. ANN or Artificial Neural Networks is one of the subsidiaries of Artificial Intelligence. The most common example of an Artificial Neural Network would be Google Lens processing and recognising the object it captures. Many of you must have used Google lens for a variety of tasks, maybe translating some text on the wall or trying to know the breed of the dog you instantly fell in love with. But have you ever wondered about the sophisticated engineering that goes behind this? We’ll find that out in this blog. 

Also Read, Resnet 50 Architecture

ANN and how it works

The best way to understand ANN and its working at the backend is to understand how you learn and perceive the surroundings around you. Remember your first time riding a bicycle? Countless number of iterations of riding, falling, and getting back up again, learning some minute details from your errors for the next time you hit the pedal. At a higher level, it is what ANN is all about. Executing a task and learning from mistakes. 

ANNs simulate the neural network in humans. A human neural network consists of billions of neurons. It’s these neurons that give humans the ability of learning and recognising things from their past experiences. These neurons work closely with each other towards a common objective. ANNs simulate this neural structure of a human brain with a layered structure, which basically consists of three major components-  The input layer, the hidden layer, and The output layer.

Source - link

The input layer

The input layer is the first component in the architecture of an ANN. It receives the input for various explanatory attributes and includes the bias term as well. So say we have n input variables or attributes, the size of the input layer will be n+1 where the extra variable is the bias term. 

Also read, Artificial Intelligence in Education

The hidden layer

This is where the majority of computation happens. It receives the processed data from the input layer. This data is nothing but the input variables. Now, It may or may not be true that all these variables are equally essential in computing the final predictions. For example, let’s say we want to train an ANN model to recognise Siberian tigers. So there are some defining features about a Siberian tiger that would be very crucial for the predictions. Like their golden fur with prominent dark stripes and the canines. These features would be given more importance over other features of the input. Now there might be relatively less crucial features like the ears or the tail. They may be used to make the prediction but wouldn’t be considered defining features of a tiger. So to differentiate between the variables based on the impact they may have on the final predictions, we assign weights to these variables. Initially, random weights are assigned but these weights as we go along. This process of adjusting the weights is called ‘Backward propagation for errors’ or more commonly known as ‘Backpropagation’. We’ll discuss this in detail later in the blog. 

The output layer

The final component in the ANN architecture. Finally, the hidden layers link to the ‘output layer‘. The output layer receives connections from hidden layers or from the input layer. It returns output corresponding to the input variables. The active nodes in the output layer combine and change values in the data to produce output values. 

The key to an effective ANN lies in the appropriate selection of weights. 

Get the tech career you deserve, faster!
Connect with our expert counsellors to understand how to hack your way to success
User rating 4.7/5
1:1 doubt support
95% placement record
Akash Pal
Senior Software Engineer
326% Hike After Job Bootcamp
Himanshu Gusain
Programmer Analyst
32 LPA After Job Bootcamp
After Job
Bootcamp

Some major processes in ANN

Forward Propagation

As the data moves from the input layer to the hidden layer and then to the output layer, computing the output for each iteration in the training phase, the process is called Forward Propagation. 

Backpropagation

We learned that we assign weights to features based on the impact they have on the final prediction. Initially, we assign random values to the weights and after each iteration( or forward propagation), the cost function is used to revise these weights. In a way, it works as a feedback system where we make our prediction which is then validated by the actual output to see how close or far the model is from the actual prediction. 

The activation function

Activation functions are mathematical equations that convert the output of each layer before it is passed to the next consecutive layer. It alters the output from the layers in a way that makes it more convenient for predictions. The most commonly used activation function is the sigmoid function which has a value between 0 and 1. 

Source - link

Advantages and Disadvantages of ANN

  1. ANN models are extremely versatile. They can be used to perform Linear and non-linear tasks alike. But this comes with its own challenges. One major challenge to ANN is the high volume of data that needs to be fed for training. It’s because an efficient model requires a variety of data to encompass all the varying input types and effectively learn for the task a model is aimed for. 
     
  2. ANNs aren’t entirely dependent on each and every unit. If a few of them fail to respond, an ANN may still be able to produce the required output. But this requires a huge computation cost, in terms of storage as well as processing power. 
     
  3. Every unit in ANNs need not be explicitly programmed for the task it is supposed to take. It’s very autonomous in task allocation and learning objectives. However, this may lead to the user being absolutely clueless about how exactly the model is making the predictions it makes. The implementation among the nodes in the network may be vague to the user himself. 

Applications of ANN

1. Speech recognition:- With deep learning, Speech recognition can be effectively implemented. Earlier we had other methods which were based on statistical data like Hidden Markov model but they weren’t as efficient as implemented by ANN. 

source-link

 

2. Facial recognition:- One of the most commonly known examples of Deep learning. Facial recognition systems can be efficiently implemented with Convolutional neural networks. Facial recognition is a very common feature in smartphones these days.

Source - link

3. Stock market prediction:- A stock market predictor can be efficiently implemented with a multilayer perceptron. MLPs consist of several layers of nodes each of which is fully connected with each node in the next layer. The input variables may be the opening price, past performance, annual returns, etc. 

 

Source - link

Frequently Asked Questions

  1. What do you understand by ANN? 
    Ans. ANN stands for Artificial Neural Network. It’s an AI computation technique that simulates the human way of thinking and cognition. The fundamental unit of a neural network is a neuron and it’s the network of these neurons that make the neural network. 
     
  2. Briefly explain Forward and backward propagation. 
    Ans. As the data moves from the input layer to the hidden layer and then to the output layer, computing the output for each iteration in the training phase, the process is called Forward Propagation. 
    After every forward propagation, the network receives the feedback according to which the weights are adjusted. This is known as Backpropagation. 
     
  3. Mention some real-life applications of ANN. 
    Ans. 
  • Facial recognition
  • Signature recognition
  • Stock market prediction

Key Takeaways

Artificial Neural Networks is a vast area of study. And it also finds its uses in a variety of day-to-day applications. This blog is a perfect starting point for beginners who look forward to learning about ANNs. The blog provides a thorough explanation of architecture and the major processes that happen in a Neural network. You may want to check out our expert-curated courses in deep learning if you want to build your skills in this domain of artificial intelligence. 

Check out this article - Padding In Convolutional Neural Network

Happy Learning

Next article
Bayesian Belief Network in Artificial Intelligence
Live masterclass