What is Deep Learning?

Machine learning & Data science with R

Looking to kick start your machine learning & data science career? Learn supervised, unsupervised machine learning models like linear regression, logistic regression, KNN, clustering, PCA, naive bayes & neural networks with this video based online course.

Do you know what is deep learning? Do you think it is same as machine learning or artificial intelligence?

I have written a post explaining differences between deep learning, machine learning & artificial intelligence.

If you are my subscriber, you must have already received the link for it, but if you are currently not subscribed to my list, click here to read it. And don’t forget to subscribe to regularly receive the updates.

In this post, I am going to talk about deep learning.

Objective is to provide a high-level information so that you understand;

  • What is deep learning?
  • Popular applications of deep learning
  • Core value or advantage of deep learning

What is deep learning?

Deep learning is subset of machine learning.

Deep learning is derivative of neural networks, also known as automated neural networks.

Automated neural networks are subset of supervised machine learning techniques. Automated neural networks are primarily used for regression and classification in machine learning, but deep learning takes this concept forward.

Deep learning models are kind of hybrid machine learning model where deep learning follow philosophy of unsupervised machine learning.

I will explain the high-level difference between supervised and unsupervised machine learning models later in this post.

But if you are interested in learning more about machine learning and ANN, you can check out my course on machine learning.

All kind of neural networks (deep learning or conventional shallow neural networks), they all try to mimic the human brain function.

For example, if you look at this diagram (I know it’s ugly but will serve our purpose), our body consist of various organs like brain, heart, hands, legs etc.

what is deep learning

Every organ depends upon brain to perform the desired or required action.

So, if you wanted to lift something using your hands, your brain will send signals to the hands and they will perform the action.

Off course, human nervous system is very complex, and people do PhD in them, so we can’t cover everything in one post but that’s how our brain and body function.

Brain sends electrical signals to various organs. And for transmission, it uses nerve cells. Also known as neurons.

There are millions of neurons in our body.

These three arrows in this diagram are representing neurons here. There will be many more neurons from brain to hands.

You can think of these neurons as wires. We use wires to connect electrical equipment like fan, TV to power outlet. These wires are used for conduction of electricity. Neurons does the same job.

Deep learning models consist of;

  • 1
    Input layer
  • 2
    Hidden layers (many hidden layers)
  • 3
    Output layer
deep neural network

Input layer consist of input parameters, which are data set’s variables or features.

Hidden layers are trying to mimic human behavior. This is where magic happens.

Output layer consist of output of the model like class.

Few examples of deep learning models (deep learning algorithms) are;

  • 1
    CNN- Convolutional neural network
  • 2
    RNN- Recurrent neural network
  • 3
    LSTM- long short-term memory
  • 4
    GAN- Generative adversarial network

Deep learning is very close to artificial intelligence as far as its application is concerned. Most popular libraries used in deep learning projects are tensorflow, pytorch.

Popular applications of deep learning

Deep learning is used in building next gen products like;

Core value or advantage of deep learning

In a typical supervised machine learning models, data scientist will decide what kind of features or variables should be used for model building. For example, if you look at this table (please ignore the values as I have manually entered everything from illustration purpose only).


Serial #

Name

Height (cm)

Weight (kgs)

BMI

FBS

Random Sugar

Age

Diabetic

1

Some name

118

88

29

100

118

45

Yes

2

Some name

93

93

33

105

93

24

No

3

Some name

102

102

31

100

102

18

Yes

4

Some name

162

62

34

98

162

55

No

5

Some name

154

54

35

91

154

45

Yes

6

Some name

176

76

31

90

176

43

No

7

Some name

177

77

25

115

177

54

Yes

8

Some name

171

71

21

125

171

66

No

9

Some name

170

70

24

256

170

71

Yes

10

Some name

168

68

28

123

168

58

No

Now we can build a predictive model using this data set using logistic regression model to predict if a person will be diabetic or not.

Serial number, names or any other unique id is not used in building machine learning models because unique values doesn’t have a pattern.

So, we will drop serial number and name from the data set while building model. This process of selecting variables or feature, is called feature engineering.

Now in this case, feature engineering is easy but if we are building a model for image recognition, it becomes close to impossible to do feature engineering.

We will have to breakdown the image to individual features like pixels and various features of the images which is impossible.

This is where deep learning comes to rescue, we can simply pass our images to the deep learning model and it will automatically detect feature.

There is no feature engineering required with deep learning models.

And this is the biggest advantage of deep learning over supervised machine learning models.

This is a very high-level overview of deep learning models.

I am working on a course on deep learning, it will include fundamentals of deep learning, convolutional neural network model development with tensorflow and pytorch. We will build a model for image recognition. This is cutting edge and very hot topic right now.

If you are interested in learning deep learning, leave your name and email so that I can send you early bird discount whenever this course is launched, and this course will launch in few weeks.

If you have any suggestion for the course or anything in general, please leave your comment.

About akhilendra

Hi, I’m Akhilendra and I write about Business Analysis, Data Science, IT & Web. Join me on Twitter, Facebook & Google+

Speak Your Mind

*