For those who aren't acquainted with the term MACHINE LEARNING, let me first give you a basic idea of it.
You're trying to make a computer smart enough to learn from the data it's fed so that after a point of time the computer is able to predict further data.
How?
Back in school we were given a table with X and Y values and made to plot the points on a graph sheet. We would then join them to make a line (assuming it was that simple).
Now given any damn X, getting a Y was a piece of cake!
This is exactly how a machine learns too. It's building a curve (or modifying it rather) every time you feed in some data so that when you ask the computer "What will be the value of Y for this X?", it's able to predict and give you the answer.
Instead of physically drawing a curve, the computer is trying to generate the equation of the curve (say Y=X - 5).
Now let's think about the practicality of this. Let's say you are making your computer predict weather.
So you'll start feeding in weather reports of every single day and every single hour of the past one year.
Now it's not just a 2 dimensional curve anymore. The temperature(z) now depends on day of the year(x) and also time of the day(y).
And going by the randomness of weather, the curve can't just be a straight line.
So the equation generated by the computer will not just have 3 variables(x,y,z) , it'll also have squares and cubes or higher powers.
As you can see, depending on the number of factors a prediction depends on and also the randomness of the outcome, the complexity of the curve goes on increasing.
Now let's think of something even more advanced. ROBOTICS!
Let's say a high-tech company has designed a robot that's capable of driving cars exactly the way you do (to make you feel more comfortable). But the catalog says you gotta have this robot sitting next to you for the first one month. The robot's gonna learn how to drive in a month!! On it's own!!
You are no more feeding values it. It's learning things on its own using its accelerometer, IR emitters and sensors, cameras etc.
Going back to our curve-logic, you have to consider 3 factors :
1) There are probably tens (or hundreds) of factors that every single decision while driving depends on.
(Amazing how our brain is doing all this without us noticing it)
2) It's a continuous process so every single millisecond, the robot is learning new decisions based on the inputs from its sensors and cameras.
So speed is a huge criteria.
3) There are going to be zillions of (factors,decision) sets generated over a period of one month that the robot has to remember. So the robot should have a huge and fast memory.
This is where THE ANSWER lies. It takes super computers which are capable of handling such volumes of data, such fast learning ability and such fast decision making based on the learning.
This is what this generation's technology is going to be all about!!
It is all the more RELEVANT in today's scenario. This is why Machine Learning is such a huge deal today!
A machine that's capable of catching hold of patterns and predicting further events based on it.
Just think of all the areas where Machine Learning can be brought in. Let your imagination run wild.
You're trying to make a computer smart enough to learn from the data it's fed so that after a point of time the computer is able to predict further data.
How?
Back in school we were given a table with X and Y values and made to plot the points on a graph sheet. We would then join them to make a line (assuming it was that simple).
Now given any damn X, getting a Y was a piece of cake!
This is exactly how a machine learns too. It's building a curve (or modifying it rather) every time you feed in some data so that when you ask the computer "What will be the value of Y for this X?", it's able to predict and give you the answer.
Instead of physically drawing a curve, the computer is trying to generate the equation of the curve (say Y=X - 5).
Now let's think about the practicality of this. Let's say you are making your computer predict weather.
So you'll start feeding in weather reports of every single day and every single hour of the past one year.
Now it's not just a 2 dimensional curve anymore. The temperature(z) now depends on day of the year(x) and also time of the day(y).
And going by the randomness of weather, the curve can't just be a straight line.
So the equation generated by the computer will not just have 3 variables(x,y,z) , it'll also have squares and cubes or higher powers.
As you can see, depending on the number of factors a prediction depends on and also the randomness of the outcome, the complexity of the curve goes on increasing.
Now let's think of something even more advanced. ROBOTICS!
Let's say a high-tech company has designed a robot that's capable of driving cars exactly the way you do (to make you feel more comfortable). But the catalog says you gotta have this robot sitting next to you for the first one month. The robot's gonna learn how to drive in a month!! On it's own!!
You are no more feeding values it. It's learning things on its own using its accelerometer, IR emitters and sensors, cameras etc.
Going back to our curve-logic, you have to consider 3 factors :
1) There are probably tens (or hundreds) of factors that every single decision while driving depends on.
(Amazing how our brain is doing all this without us noticing it)
2) It's a continuous process so every single millisecond, the robot is learning new decisions based on the inputs from its sensors and cameras.
So speed is a huge criteria.
3) There are going to be zillions of (factors,decision) sets generated over a period of one month that the robot has to remember. So the robot should have a huge and fast memory.
This is where THE ANSWER lies. It takes super computers which are capable of handling such volumes of data, such fast learning ability and such fast decision making based on the learning.
This is what this generation's technology is going to be all about!!
It is all the more RELEVANT in today's scenario. This is why Machine Learning is such a huge deal today!
A machine that's capable of catching hold of patterns and predicting further events based on it.
Just think of all the areas where Machine Learning can be brought in. Let your imagination run wild.
No comments:
Post a Comment