'I wake up in the morning. As soon as I unlock my phone, a Google card tells me my ETA to work.
Heading to work, I always listen to some music on Spotify. All I need to do is turn it on, and it generates a playlist, with exactly the kind of music that I like.
At work, I'm supposed to decide which stocks we hold on to, and which ones do we sell. My team has their experience, and equally importantly, a software which analyses the market and generates handy insights for us.
Back from a long day, I browse Facebook for a while, looking at the dankest memes of the day. In between them, I find an ad for a cool T-Shirt just like the one I ordered yesterday.'
Given above is the average life cycle of a corporate. What is common in all the things mentioned above?
ALL of them employ MACHINE LEARNING.
Machine Learning has such a broad horizon in the world of today that it is seemingly impossible not to encounter it. Literally all spheres of life are flooded with the applications of this technology, and rightly so. Machine Learning has given humans to a great degree, what we have always yearned for; predictability.
Evolution of Machine Learning
The concept of Machine Learning is not entirely new. It was in the 1950s when Alan Turing created the Turing Test, predicting that a day will come when you would actually have to differentiate between a computer and human.
Turns out, he wasn't wrong (Read: Eugene Goostman, the bot which beat the Turing Test).
In 1952, Arthur Samuel wrote a program which beat him at checkers, creating a history which would be repeated almost 50 years later when World Chess Champion Gary Kasparov was beaten by IBM's Deep Blue.
It was in the 90s that Machine Learning faced a huge shift. Focus changed from knowledge based approach to a data driven approach. People realized that if computers were given enough data, they could use it to analyse conditions previously not encountered. In short, this is when Computers actually started learning.
Today, Machine Learning has conquered many more domains with interesting applications such as Face Recognition, Sentence formation, etc. with notable players like Google, Facebook and Amazon leading the race. (Read: What do the Big 4 think about AI)
What exactly is Machine Learning?
Enough with the build-up. In short:
Machine Learning is making a system perform a task that it hasn't been specifically coded to, enabling the system to make decisions based on the data it learns from.
What does that mean?
Take an example: Suppose I intend to create a spam classifier. My program should take in an E-Mail, and it would output whether the mail is Spam or not.
A very simple approach to tackle this problem would be to check the email for some words which usually occur in Spam mail (such as 'free') and make a prediction based on that. The general code for such a method would look like this.
If(mail contains words like 'free')
print("Spam")
else
print("Not Spam")
Now, this method suffers from a lot of limitations. First and foremost, not all mails containing the word 'free' must be spam (A loved one may be asking whether you're free for a meet-up). Hence, we should be checking for the context in which the word is encountered and the words present along with it, rather than naively looking for that word. Also, it is very difficult to manually define all words which go towards characterizing a mail as spam.
What a Machine Learning algorithm would do in this case is that it would read through a lot of spam and non-spam mails, assigning different weights of importance to each word or a sequence of words. So when it encounters a new mail, it reads through the words, sums up their weights. If this adds upto a value above a certain threshold (that you have decided manually or left to the program) the mail is probably spam. Otherwise, it is not.
You may ask, 'That's cool, but why can't I hard code this?'
You can, if the dataset is small enough. But the sheer size of actual datasets makes this virtually impossible. Hence, let ML do the heavy lifting.
Importance of Machine Learning
What makes Machine Learning amazing is that it can be employed in so many different fields which may not even be remotely related to Computer Science, and yet be able to deliver solutions that vastly optimize the business. This gives them a huge advantage over competitors who stick to conventional methods and 'intuition'.
Given the fact that billions of Gigabytes of data is generated everyday around the globe (not kidding), some of the most prominent aspects of ML are Data Mining, Big data and Data Warehousing. Imagine the amount of hard coding that would be required to generate patterns and analyse this (Hint: Not a good Idea). This is where Machine Learning comes into play. Just get the data to be refined enough so that interesting results can be obtained.
Scope of Machine Learning
The vast domain of ML needs to be categorized into different sections for understanding. Here we look at some of the prominent parts of Machine Learning:
1. Natural Language Processing:
Imagine a system which automatically creates a summary for your text, or a system which can learn and predict linguistic constructs. This is what Natural Language Processing, often shortened as NLP, is about; making the machine learn to work with languages that humans speak, like English, French, Hindi etc. by providing them with experience in the form of datasets.
There are Machine Learning projects involving NLP, which can even create stories by reading and analysing sufficient data. Now that's cool, isn't it?
2. Digital Image Processing:
Probably the best example of Digital Image Processing is Google's self-driving car. Another famous application of DIP is Facebook's face tagging system, and Google Photo's face recognition system by which it automatically creates folders for your friends' and family members' photos, by identifying the similarity in the pictures.
3. Digital Signal Processing:
The virtual assistants like Siri, Google Assistant, Cortana, and Alexa, which have become an integral part of our digital environment, all work on speech processing, which is derived from Digital Signal Processing and Natural Language Processing. Same goes for the Chatbots, and systems which identify song's genres.
DSP analyses the waveforms of the given discrete signal and classifies the clip based on that (after all, that's what all computer systems understand, right!).
What isn't Machine Learning?
Is AI and Machine Learning the same thing?
Well to be very frank, the simple answer is no. They aren't the same, but yes, they are inter-related terms. AI systems are basically smart bots, which can think for themselves like humans do, can have their own communication method(Read: Facebook shuts down AI after it starts communicating in a new language), and can react to external stimulus. Whereas Machine Learning systems are those which can adapt patterns from datasets and can be used as either predictive or as discriminative models. They need not be hard coded to play a game or segregate huge volumes of data, but they learn from experience, i.e. by analysing data.
A good way to imagine the relationship between these two is that Machine Learning is the model through which Artificial Intelligence is implemented, i.e. basically Machine Learning powers AI systems.
Where does deep learning come into picture?
Deep learning is basically a more specific implementation of Machine Learning, where we use a structure called an Artificial Neural Network (ANN). While Neural Networks have been around for long, it is only now that we have enough computational power to actually make networks dense enough to be useful. Now, coupled with the power of GPUs, and over-abundance of data, Deep learning has reached a very high degree of precision.
What makes deep learning so important is that, with this much power, deep learning has the potential to make all conventional ML algorithms obsolete. The concept has been a disruptor at all ML conferences recently. The downside of this is, Deep Learning is available to only those who have enough funds to implement it, while ML had more of a mass appeal.
Moving Forward:
By now, we have a basic idea of what Machine Learning is all about, its domains, scope and application.
But what are the kinds of machine learning algorithms, which make it such a powerful tool? More importantly, how do we apply them? What are the set of steps to get the job done?
On this happy note, we leave you with this knowledge, and the questions posed above.
Stay tuned for our next article on Introduction to Machine Learning.
Thank you, and may the code be with you!