How Machine Learning Works - Mathematics every where

Hey Readers !!, Welcome back, Hope you have read the previous post before we start with this. Check out the previous post on How Machine Learning works Part 1.

For those who already read my previous post, Thanks a lot for continuing the journey to find the basics of Machine Learning. We stopped the previous post saying that we find the trends / Patterns that occur in the data and that forms the basis for the Machine Learning.

Here we are starting with the topic to understand FROM WHERE DOES THIS MATHEMATICS came into picture for machine learning.

Mathematics every where

There was a statement in the previous post that all things in the nature follows some sort of Mathematics in the form of any physical science concepts. In the same way the patterns that exist in the nature that are visible in the data also rely on Mathematical concepts or the patterns can be represented in the mathematical form through the guiding equations. 

y = mx + c is a mathematical form of the Linear equation. 

On seeing the applications of Mathematics in the machine learning, we can see that its the basis for forming a model, it also helps the data to pre-process or helps in arriving to the final state by some transformations. By the word Arriving to the final state means the scenarios in which the final state cannot be reached until we convert the existing problem to the different angle or plane or environment and solve it there and brings back. 

For example, Imagine more there is a table of data with 20 columns, we call each column in the table as an attribute or feature or dimensions. We try to create a model means the guiding equation that represents the co-relations / patterns in the data. Imagine our developer doesn't have a good processor or GPU to do that much volume of data, we minimize the running time by mapping the 20 dimensional data onto a plane which is of 2 dimensional and this is easy for the processor to come to the final intermediate state and the final results that are on the 2D plane are mapped back to the 20 Dimensional plane. 

Don't get confused with the thoughts on how this can be possible or How it can be done ?, We keep it simple and there are lot of techniques like this. All theses techniques helps us do a processing faster or to arrive at the end faster. For all these Mathematics is the guiding factor.

I can hear that your mind says OMG, I should have taken Mathematics paper in schooling well !!.

Skill and Areas:

When we speak about what skill a Machine learning person should posses, we see think of the following,

1.  ................................
2. Python Programming :-)
3. Analytical skills 
4. Learning Various Algorithms...
5. Hard work :-P & Business Understanding Capabilities..etc

Purposefully the 1. ...... is left blank, because the candidate skill that is gonna fill that slot is ranking one what ever skill rest be.

The candidate skill is " THE ABILITY TO VISUALIZE THE PROBLEM AND TRANSFORMING THE SAME TO MATHEMATICAL DOMAINS.

Yes, being a Machine Learning professional, to solve a problem in the world we should have the sens of converting the problem into the mathematical sense. For example, Take any problem not needed to be the one that can be solved by machine learning or not like below,

John's doctor asked not to stay in the very high areas may be at the height of 100 feet and above, as he will suffer from breathing issues. John planned to buy the home in a flat and he is about to book it. He needs to decide on which floor he needs the home. He can't go to each and every floor and experience if have any issues in breathing, but he believes in doctor and decided to stay below 100 feet of height.

How to solve this Mathematically,

Any standard floor will have an average of 10 feet of height, doctor said the problem is above 100 feet, that means that he can stay below 9th floor because the floor starts with the ground floor its not 10th floor but the 9th floor. For additional safety he left the 8th floor too and wanted to book a home in 7th floor because in 7th floor the height at which he would live is 70 feet average, which is on a safe side.

The above example could be a simple one, but there are situations in the world where there are complex problems that rely only on mathematics like launching of a satellite to the orbit. So one should have the ability to visualize the problem in the mathematical domain.

Domains that guide us in Machine Learning

We cannot claim that only the following domains would be used in the machine learning, as of now theses are the most used one, If you could reach the level of research you can invent a Machine Learning Algorithm that will even use a Pythagoras Equation too.

1. Statistics - The most and foremost important field
2. Integral and Differential Calculus - Needed to find the optimizations in the model deriving and processing.
3. Transformations - Needed to transform the data and solve the problem.
4. Algebra - To create the models.
5. Predicate Calculus - Helps in rule based models like recommendation systems.
6. Probability - The second foremost field than others to arrive at the conclusion that my model works better. Any model cannot be said its the best, but better than before because of the rule that machine learning says as "Something in place of nothing".
7. Lot more fields as per the advanced applications .... !! 

We are mainly discussing about the "HOW MACHINE LEARNING WORK?", from the last and the current blog. Here in this blog we can understand any thing in the machine learning depends on the Mathematics to the core.

Does this answers our question "HOW MACHINE LEARNING WORKS ? ", NO, Not yet, we need to see some more to get the complete understanding of how it works

Its a basic tendency of human, to keep the no-longer than 20 minutes solid on a stuff, So I am splitting the big topic into one more topic as below,

How Machine Learning works? - Data rules the kingdom