Machine-Learning 0 to 1 - Article 1

Hi Guys,

Last week, i have announced my first series on machine learning. today we are starting with first article on machine learning.

In this first article we will learn,

1. What is machine learning
2. How machine learning works
3. How Machine learning different than programmatic approach, and we will see what is learning
4. Types of machine learning algorithms
5. Some real life examples of machine learning
6. Let’s begin with small introduction on Machine learning.

As we all know, AI & ML is one of the hottest topic of 21st century. Everyone is talking about, and people are using it nicely in almost every domain.

But,

Do you know when machine learning comes ?
Can you guess?
In 2000 or In 1950 or even earlier in 1900 ?
You guys won't believe it comes since mid of 17th century. You can check history is ML here

Basic idea behind research of machine of learning was to make machine intelligent as human.

We all knows how new born baby become smart human with time. In fact we all learn through reading and our experience, as we grow. Or you can say human has capability of self learning.

Humans can do self learning through past experience & his knowledge.

Until ML come into picture, computer is able to learn only through hard coded program. It mean whatever you want to achieve, you need to write program for it, and on the basis of which computer able to performed.

In short computer were only follow instruction we provided. It didn’t able to learn by itself. Self learning is missing for computer like human have.

Machine learning overcome self learning problem.

What is ML ?

Machine learning is responsible to make computer as intelligent as human, and enables it to self-learn without being explicitly programmed.

How ML works ?

So let’s talk about how machine learning works, as we already know one can only do self-learning from his past experience only. we need to understand how computer can learn from his past experience ? so researcher find a way to do it. They have created data set(knowledge set) from past experience and write some programs / techniques to learn from that data. Those programs are nothing but algorithms.

Currently there are plenty of machine learning algorithms available in market. And researcher are continuously doing research on it. We will cover some core algorithms in these series which are most useful now a days.

let's see how ML is different than programmatic approach.

Programmatic Vs Machine Learning Solution

In Programmatic solution, you need to give program and input data to your computer, so computer will use your program to generate output.

But in Machine learning approach, you need to give sample (input/output) data to computer as well as your input data(for which you want output). And computer will generate program or you can say Model as a output. And you can use that Model to solve subsequent task.

Let’s first understand what is Learning,

Learning is the ability to improve once behaviour with experience. In short build a computer system which improve with experience.

let's check former definition of Machine learning as given by Tom Mitchell

A computer program is said to learn from experience E, with respect to some class of task T and performance measure P. If it’s performance on task in T, as measure by P improves with experience E

T - Task (like Prediction, Classification..)
E - Experience also called sample data.
P - Performance measurement. Let say you want to increase accuracy in prediction / problem solving. Corresponding to this you can define the Performance measure P.

Based on this definition we can look at learning system as a box, to which we feed the experience or the data (E), and there is a problem or a task (T) that require solution. (we will also give background knowledge which will help the system) and this problem/ Task learning program comes up with Model or solution, and its corresponding performance can be measure.

Below is the semantic diagram of a ML system.

Inside the black box, there are two main components

Leaner:
It takes experience/data and background knowledge, and build the models

Reasoner:
It use that Model built by leaner, with given a task find the solutions to the task

Steps to create a learner:

1. Choose/Prepare the training data
2. Choose target function, how we want to represent the Model. This what we want to learn (For example if we write to try Machine learning system to play game of checkers, The target function would be given a board position what move to take)
3. Choose how to represent the target function (linear / decision tree / or something else)
4. Choose learning algorithm (which we will going to learn in next articles)
5. First & third steps are the most important step in designing of a learning algorithm.

Let’s take a look into one example of machine learning in details. As we already discussed ML is used in almost all domain. But let’s take a example of “diagnose a disease”

Input: symptoms, lab measurement, test result, DNA tests etc..
Output: one of the set of possible diseases or “none disease”

For doing this one can data mine historical medical record to learn which future patients will respond best to which treatments.

There are mainly 4 types of machine learning algorithms as below -

1. Supervised algorithm
2. Unsupervised algorithm
3. Semi-Supervised algorithm
4. Reinforcement algorithm

We will look each type of algorithms in detail in next part of this series.

Is machine learning magic ?

Once you start seeing how easily machine learning techniques can be applied to problems that seem really hard (like handwriting recognition), you start to get the feeling that you could use machine learning to solve any problem and get an answer as long as you have enough data. Just feed in the data and watch the computer magically figure out the equation that fits the data!

So remember, if a human expert couldn’t use the data to solve the problem manually, a computer probably won’t be able to either. Instead, focus on problems where a human could solve the problem, but where it would be great if a computer could solve it much more quickly.

Real life examples Of Machine Learning

• E-commerce giant like Amazon using ML to recommended products on the basis of user’s purchasing pattern
• Facebook using ML to automatic recognize your friend’s face and ask you to tag them
• Uber using ML to estimate time from source to destination
• Google use ML in many ways like,
  • in google maps to extract street names and house number from photo taken by street view cars,
  • In gmail to detect spam email
  • In youtube to recommended videos from your watching pattern
• Bank are using ML to detect fraud

These are basic examples, but in today’s life we are using many machine learning applications daily and we even don’t know. Try to think about all the app you are using. 70-80% of them are using ML. For example Gmail, uber, facebook, twitter, etc...

Personal note for newbee:

ML is not like other technologies, where you can just read theory and you can able to use it. If you want to learn ML in a right path, try to discover different problem and think about it’s solution. Because by knowing theory only, you can not become master in ML. so if you want to be a master in ML do practical more rather than reading. So try to solve as many problem you can.

Next week i will come up with new article on
Types of machine learning algorithms in which we will see different types of algorithms available, which algorithm use in which condition, real-life examples etc.

Next couple of weeks will be fantastic for both of us, stay in touch guys.

Thanks for all your support in advance.

-Hemang