Well now a days everyone seems to be talking about machine learning and its applications/uses, but have we ever thought how all of a sudden ML has become so popular? If I tell you that work on AI started way back in 1950 and Machine learning started to grow rapidly in 1990, what has suddenly given a boost to Machine Learning?
In this blog, I will give you answers to these questions but let us first have a look at what machine learning is.
We will start from basics and understand what a Program is.In simple terms,a program is predefined set of rules or instructions. When data is fed to the computer, it processes the data using these rules. That sounds pretty cool, but then came this question of can’t a computer be just fed with the data and it decides rules and give us the answers. This would make our jobs easier as the computer would decide what to do with the data rather than the developer modelling the data and defining set of rules.
This notion of computer deciding set of rules to be applied on data gave rise to Machine Learning. The machine would figure out the rules and when we input the data it would give us the answers.
In Machine Learning rather than writing programs, we train the machines explicitly. It is presented with historical data and it tries to find patterns in that data. It can do it by finding underlying functions that represent the data or we can just provide it with historical data and also do some tagging/labeling e.g. for each row that we feed we provide the outcome/answer.
Let us take the example of finding whether an email is spam or not. In this case we take all the emails and label them as spam/not spam. Then we feed this data to the machine. Now that the machine has learned the pattern from historical data, it will apply it to a new email and predict it as spam/not spam.
How the machine does this is by using different algorithms which try to find patterns in data.
Now, that you know what machine learning is, let us have a look at the reasons why ML has suddenly become so popular:-
1. COMPUTING POWER
ML is gaining a lot of popularity because of exponential growth in computing power. We have seen almost 1 trillion fold increase in computing power from 1956. The fact that a single apple iphone 5 has 2.7 times the processing power that supercomputer Cray-2 had in 1985 speaks a lot about advances made in field of computing power.
ML uses statistical methods/algorithms to find functions or patterns that give data that structure. Performing this analysis on large
datasets was computationally not possible earlier. But as the computation power grew multifolds, performing these operations became possible.
2. LARGE DATA SETS
With the invent of Internet, IOT, and Big data, the data growth has been on constant rise. The data volumes are exploding so rapidly that more data has been created in past 2 years than in the previous history of mankind. Therefore, now we have large data sets which computers can analyze to find a lot of hidden patterns,features.
3. EMERGENCE OF STATISTICAL LANGUAGES
One more important factor responsible is emergence of statistical languages like Python, R, SAS which make it pretty easy to perform machine learning and other AI related development activities. Python though is the most favored among all, because of its simplicity. Syntaxes in python are very simple and it takes short development time compared to other languages. It supports object oriented, functional as well as procedure oriented style of programming. To add to this we have plenty of libraries in python that make our tasks easier.
4. ADVANCEMENT IN ALGORITHMS/PACKAGES
There is lot of advancement in algorithms as well. While neural networks have been used from a long time, we now have more efficient way of teaching individual layers of neurons known as “deep learning”. The first layer learns primitive features and once it correctly recognizes those features, they are then fed to next layer which trains itself to recognize more complex features and so on. This process is repeated until the system can recognize the object. These deep learning networks have further improved the accuracy and taken out any human intervention which was required in case of supervised and unsupervised learning.
Thanks for reading and do let me know your views in comments.
