Introduction to Deep Learning

Overview of Deep Learning

From the moment we open our eyes in the morning, our brain starts collecting data from different sources. To keep up with the pervasive growth of data from different sources mankind was introduced to modern Data-Driven Technologies like Artificial Intelligence, Machine Learning, Deep Learning etc. These technologies have engineered our society in many aspects already and will continue to do so.
This tutorial series guides you through the basics of Deep Learning, setting up an environment in your system to building the very first Deep Neural Network model.

Table of Content:

• What is Deep Learning?
• Why do we need Deep Learning?
• Importance of Deep Learning
• Applications of Deep Learning:
• Deep Learning vs Machine learning
• Why should you opt for Deep Learning now?
• Biological Neural Network vs Artificial Neural Network:
• How Do Neural Networks Work in Deep learning?
• Perceptron
• Deep Neural Network:
• Conclusion

Watch this Deep Learning Tutorial

Deep Learning is a subset of Machine Learning which is used to achieve Artificial Intelligence. Confusing? Let us look at the diagram given below to have a better understanding of these words.

In other words, Deep Learning is an approach to learning where we can make a machine imitate the network of neurons in a human brain. It consists of algorithms that allow machines to train to perform tasks like speech, image recognition, and Natural Language Processing. It is a statistical approach based on Deep Networks, where we break down a task and distribute it into machine learning algorithms. These algorithms are constructed with connected layers. In between the first layer or input layer and the last layer or output layer we have a set of hidden layers in between that eventually gave rise to the word Deep which means networks that join neurons in more than two layers. These neurons are connected to one another, which propagates the input signal after it goes through the process. In Deep Learning a network can consume a large amount of input data, then process them through multiple layers because of which we can learn complex features of the data.
Now that we have gathered an idea of what Deep Learning is, let’s see why we need Deep Learning.

Interested in learning Machine Learning? Click here to learn more in this Machine Learning Training in New York!

Our human brain can easily differentiate between a cat and a dog. But how can we make a machine differentiate between a cat and a dog? We would train the machine with a lot of images of cats and dogs. Then once the training is done we will provide the machine with an image of either cat or a dog. Now, we will manually extract some features from the image annotation and make a machine learning model out of it, which would help the machine recognize the input image. And then the machine learning model will predict whether it was a dog or cat. It was easy, wasn’t it? But what will happen when we have a large number of inputs? Manual extraction of features for a large input is backbreaking work.
What if we could skip the manual extraction part? Wouldn’t it make things a lot easier? When the amount of input data is increased, traditional machine learning techniques are insufficient in terms of performance. That is when Deep Learning came into the picture.

The following points stress on the importance of Deep Learning:

• The need for an increase in the response time is a very important matter of concern for the majority of companies and it helps in achieving this and promoting productivity and efficiency.
• It is absolutely essential for conducting effective market analysis.
• Deep Learning techniques can be applied to Big Data to yield better results by facilitating analytical and interpretative capabilities. A more personalized solution can be given to complex business problems.
• The use of Deep Learning also helps in making easy predictions and identifying trends and patterns.
• It also finds its application and usage in the domain of cybersecurity.
• Organizations these days are focusing on building models using deep learning for optimization of their logistic systems and processes.

Become a Master of Machine Learning by going through this online Machine Learning Course in Hyderabad.

There are various applications of deep learning are:

Healthcare:

Deep Learning and its innovations are advancing the future of precision medicine and health management. Breast Cancer, Skin Cancer diagnostics are just a few examples of Deep Learning in Health Care. In the coming years computer-aided diagnosis will play a major role in healthcare.

Computer vision and pattern recognition:

Describing photos, restoring pixels, restoring colors in B&W photos and videos.

Computer games, robots & self-driving cars:

Self-driving cars, beating people in computer games, making robots act like human are all possible due to AI and Deep Learning.

Voice-activated intelligent assistants:

Apple’s Siri, Google Now, Microsoft Cortana are a few examples of deep learning is voice search & voice-activated intelligent assistants.

Advertising:

Deep Learning makes allows and publishers and ad networks to leverage their content to create data-driven predictive advertising, precisely targeted advertising and much more.

Predicting Natural Calamities

Predicting natural hazards and seating up a deep-learning-based emergency alert system is to play an important role in coming years.

Finance

Analyze trading strategy, review commercial loans and form contracts, cyber-attacks are examples of Deep Learning in the Finance Industry.

Both the terms- Deep Learning and Machine Learning are used interchangeably on many occasions. But they are different from each other in many ways. Below we have discussed the most important basis of difference between the two:

Basis	Deep Learning	Machine Learning
Definition	Deep Learning is a type of Machine Learning itself but involving artificial neural networks and more algorithms.	Machine Learning means building and using systems that have the ability to learn and adapt to human instructions through statistical models and tools.
Subset	It is a subset of Machine Learning	Machine Learning is a subset of Artificial Intelligence.
Purpose	The sole purpose of Deep Learning is to solve complex Machine Learning issues.	Machine Learning focuses on developing a computer system that can learn and adapt on its own from the data being fed to it.
Type of Data used	The data used in Deep Learning is usually consisting of neural networks (ANN).	Machine Learning uses data that is mostly structured data.
Evolution	Going by the processes and techniques, Deep Learning is basically a means to explore the deeper possibilities of Machine Learning and hence is clearly an evolution of ML.	However, considering the processes of machine learning on a broader level, it is clearly an evolution of Artificial Intelligence.
Form of Outputs	The output in Deep Learning processes is usually in the form of numerical values to free-form elements.	Machine Learning output format is usually in the form of numerical values.
Time involved	It involves a huge size of data points that makes it time consuming.	On the contrary, the size of data points in machine learning is small enough because of which the models take less time.
Feature Engineering	In Deep Learning, the neural networks have the ability to automatically detect the features and due to this Feature Engineering is not required.	Feature Engineering is highly required in Machine Learning.

• Pervasive growth of Data and collection of Data became easier.
• Advancement of modern hardware and software technologies helping us benefit from the massive data.

We have both collection and access to the data, we have software’s like TensorFlow which makes building and deploying models easy. That is how Deep Learning is reshaping automation industry in a big way, becoming one of the hottest evolving technologies of 21^st century. Which also means that this is the perfect time to acquire this skill.
So now that we have learnt the importance and applications of Deep Learning let’s go ahead and see workings of Deep Learning. Also, we will discuss one use case on Deep Learning by the end of this tutorial.

Read about the major implications of Deep Learning technology in our detailed blog on the Importance of Deep Learning.

Before moving ahead with how Deep Learning works, let us try to understand take how biological neural network works.

Our human brain is a neural network, which is full of neurons and each neuron is connected to multiple neurons. Again, neurons have several Dendrites. Dendrites collect input signals which are summed up in the Cell body and later are transmitted to next neuron through Axon.
Similarly, in an artificial neural network, a perceptron receives multiple inputs which are then processed through functions to get an output. But in case of artificial neural network weights are assigned to various neurons. Then in final layer everything is put together to come up with an answer.
Let us compare Biological Neural Network to Artificial Neural Network:

Biological	Artificial
Dendrites	Inputs
Call Nucleus	Nodes
Synapse	Weights
Axon	Outputs

Read our detailed blog on Deep Learning Interview Questions that will help you to crack your next job interview.

The functions in Deep Learning are based on nodes. These nodes resemble the neurons which constitute the human brain. In the deep learning mechanism, thousands of signals travel from one node to another and assign the weights corresponding to the particular nodes respectively. Whichever node has heavier weight assigned will have a greater effect on the subsequent node and the layer of nodes and so on. When this process reaches the final layer of nodes, the weighted inputs get converted into an output. While the input data gets processed in this entire mechanism, the artificial neural networks classify it on the basis of answers received from a series of binary true/false questions. Also, deep learning systems need high power hardware as the processing of large amounts of complex data sets is needed along with advanced mathematical calculations. In a given period of time, the neural network model trains itself and with every growing stage of learning, the probability of the right answer increases. For example- consider a facial recognition system that takes some time to learn on its own but once it has trained itself completely, the probability of the right face recognition increases.

Enroll in this Advanced Computer Vision and Deep Learning course to enhance your Career!

A perceptron is an artificial neuron unit in a neural network. It is an algorithm that enables neurons to learn and processes elements in the training set one at a time for supervised learning of binary classifiers that does certain computations to detect features or business intelligence in the input data.
There are two types of Perceptrons:

• Single Layer Perceptron and
• Multilayer Layer Perceptron.

Single Layer Perceptron:

Single layer Perceptrons is the simplest type of artificial neural network can learn only linearly separable patterns. This type of perceptron is based on a threshold transfer function.

For the best of career growth, check out Intellipaat’s Machine Learning Course and get certified.

Multi-Layer Perceptron:

Neural networks with two or more layers are called multi-layer perceptron. This type of neural network has greater processing power. In this, the algorithm consists of two phases: the forward phase where the activations are propagated from the input to the output layer, and the backward phase, where the error between the observed actual and the requested nominal value in the output layer is propagated backwards to modify the weights and bias values.

Thinking of getting a master’s degree in AI? Enroll in Master’s in Artificial Intelligence in India.

Deep neural network refers to neural networks with multiple hidden layers and multiple non-linear transformations.

As we can see above, simple neural network has only one hidden layer, whereas deep learning neural network has multiple hidden layers.
Understanding workings of Deep Learning with an example:
Here we are going to take an example of one of the open datasets for Deep Learning every Data Scientists should work on, MNIST- a dataset of handwritten digits. This is one of the most popular deep learning datasets available on the internet.

About MNIST:

• It has 70,000 images in 10 classes (0 to 9)
• Out of those 70,000 images, 60,000- training set and 10,000-test set.

Go through this Artificial Intelligence Interview Questions And Answers to excel in your Artificial Intelligence Interview.

Working Explanation:

• Consider neurons as something that hold a number between 0 to 1, called activation In our case it represents the grey scale value of the corresponding pixel.

• Each one of these images consists of 28 x 28 pixels=784 pixels.

• These 784 neurons form our first layere. the input layer of the network.
• Output layer has 10 neurons with activation number from 0 to 1 representing digits from 0 to 9.
• Feed hidden layers an image with an activation function, which causes a specific pattern in the next layer, which causes another pattern in the one next to that.
• Finally, we get some pattern at the output layer as well.
• Neuron with the highest activation i.e. the brightest one is the output of the network.

Now, let me ask you a question, what role do the hidden layers play in this process? To understand that let us relate to the biological neural network system and how our brain would recognize a digit from an image.
When we see an image of the digit 9, our brain breaks it down as one circle on top. And one line on bottom. Which separately represents 0 and 1. Similarly with 8, one circle on top another on bottom.


Similarly, in deep learning, hidden layers break down the components of the given image forming a pattern. Feed in the image of 9, some specific neurons whose activation would become close to 1.

Combination of these components will trigger a neuron(see the last neuron of the output layer ) with high activation in the last layer. Thus, giving us an output digit.

Deep Learning Platforms:

Some of the well-known platforms for Deep Learning:

• TensorFlow
• Keras
• Torch
• DL4J

In this tutorial series, we will be focusing on modelling our very first Deep Neural Network using TensorFlow. TensorFlow is one of the best libraries available to implement deep learning. TensorFlow is a software library for numerical computation of mathematical expression.
Next part of this tutorial guides you through the basics of TensorFlow and its installation  on your system and how tensor flow helps us implement Deep Learning.  Jump right into the TensorFlow Use Case Tutorial, if TensorFlow is already installed in your system.

To get a more elaborate idea of the algorithms of deep learning refers to our AI Course.

The domain of Artificial Intelligence and Machine Learning is constantly evolving and so is Deep Learning. In this blog, we explored the various concepts associated with deep learning such as neural networks, neurons, perceptrons etc. We also understood the need and importance of Deep Learning in this era and how organizations are using it. We also looked upon the applications of Deep Learning and the differences between Machine Learning and Deep Learning. There has been a lot of discussion lately regarding the advancements in Deep Learning and it is definitely true that this domain is going to expand further, thus generating more career opportunities. We hope this blog will be of help to you in your career.