How to switch from Machine Learning to Deep Learning in 5 steps ?

One of the best questions of majority data science aspirants or intermediate data science practicer . Really Deep Learning seems more harder than traditional machine learning. But the truth is if you the basics of deep learning , this transition will be easy for you . My inbox is full of such questions hence today we ( Data Science Learner Team ) decided to create an overview article on – How to switch from Machine Learning to Deep Learning in 5 steps ?

Machine Learning to Deep Learning in 5 steps-

Step 1 – ( Why deep learning ?)

Understand why you need deep leaning over machine learning . See it is not always to require to use deep learning over machine learning . There are only some specific cases when it(Deep Learning ) is best to have or must to have technique .See ! there are some factors which effect this decision making – Size of data , Hardware , Feature Engineering and Execution time etc .If you have some similar concern as we mention above , It is good to go with deep learning otherwise machine learning is best for you .We have a separate article on it . Please go through the article – Difference Between Deep Learning and Machine Learning .

Step 2 – ( Basic understanding for Deep Learning Concept)-

Deep learning is completely on neural networks . Hence you should understand the basic concept of these neural network like gradient descent , loss function , activation function , back propagation , hidden layers etc . Once you understand them it will be easier for you to understand the implementation of ANN , CNN, RNN etc for you .

Step 3 – ( Implement using some high level deep learning framework – keras etc )-

I am telling you that keras has very easy implementation for ann , cnn and rnns .Basically it will give you same experience as Scikit learn gives in machine Learning .   You may create your very complex ann network in less that 25 lines of code . I know it is difficult to believe right ? See here only Model creation is not everything but you need to care of its deployment and performance  stuffs .

Few things are important to answer –

  1. How will you measure model’s performance ?
  2. How will you tune and improve its  this deep learning model ?
  3. How to decide the numbers of hidden layers in deep learning networks ?
  4. Which loss function is perfect for model evaluation etc ?

Step -4 ( How to make training on going in Production ? ) –

Most of the project remains in POC only . They never reach to production level . So most of us never explore this phase in starting few years in Data Science . This is really a good point to thought . Specially from the testing point  of views if it is a integrated part of a product . Where its outcome is next pipeline and results into some actions .

Step -5 ( Use low level implementation ) –

There are so may low level framework for deep learning like tensor flow . These frameworks provide a good control while model development . Obviously it is difficult to implement these but once you understand the concept and basics , It is easy to go with it also . Only some line of code increases .