My understanding of Machine Learning & Deep Learning

Source: Deep Learning on Medium


Go to the profile of Rahul Dangi

-Naive explanation for beginners

Credit: Hackernoon

This post is for people who are beginners in ML (Machine Learning) & DL (Deep Learning). In this article I will answer following questions:

  1. What is ML and DL ?
  2. Differences in ML & DL?
  3. What are the prerequisites for ML & DL?
  4. Who should Learn ML & DL?

What is ML and DL ?

Function is mapping in which the output will be dependent on input(call it X) and other parameters(lets call them W (weights)). For example:

y\ =\ w_{1} x\ +\ w_{0}; input is x and w_1 and w_0 are parameters

y\ =\ \frac{1}{1+e^{( w_{1} x+w0)}}; input is x and w_1 & w_0 are parameters

Why I started with a function?

We want a function which can output as per our inputs, in which many {input,output} pair is given to us and another set of {input} we have, for we need to find {output} correspondingly.

ML and DL is that function which maps outputs with the inputs correspondingly.

How to find that function?

To find such function we have two freedoms:

  1. Any combination of W (weights) and X (inputs)
  2. What are the values of W

The above first step in ML & DL terminology is called model and above step 2 is called training.

There are some standard models in ML like Perceptron, Logistic Regression.In DL we have Neural Networks as our model.

Step 2, that if finding the values of ‘W’. Once we have the model finding the values of weights is done by training. Training is iterative algorithm, in which after every iteration we have some set of W. We want such combination of weights, that the predicted and actual {outputs} are very close which is checked by cost function.

Don’t worry cost function is just another function to check the closeness of actual and predicted outputs. For more details check link :)

Once you have the model with trained weights, done.

Differences in ML & DL?

These differences are not hard & fast and are mentioned based on observations.

  1. In ML we have mostly standard models which perform very well whereas in DL we have very complex Neural Networks in which we have freedom of layers, number of neurons.
  2. ML models need features as input whereas DL models are capable of extracting the feature and needs raw data as input.
  3. ML models need less data then DL; DL models are compute intense beacuse of model complexity and large data.
  4. ML models canbe trained on CPUs whereas DL models mostly require GPUs for training.

Prerequisites for ML/DL?

If you want to learn ML/DL then knowledge of following is required:

  1. Linear Algebra & Differencial equations
  2. Probability & stochastic process

Apart from that Python will be the language that is preferred because of abundance of libraries for ML/ DL.

Who should Learn ML & DL?

If you following the latest trends in technology then I don’t need to convince you to learn ML/DL. Today due to availability of gadgets a lot of data is generated which can be used efficiently to improve your business from recruitement to marketing.

Industrial revolution, which took away manual jobs and now its Artificial Intelligence. It started already like chatbots, self driven cars and very long list.