When Deepmind’s AlphaGo won 4 out of 5 matches against the then Go Champion Lee Se-dol, the entire world took notice. Heralded as a triumph for Artificial Intelligence, this moment was an important step towards making sophisticated AI powered machines a reality.
This was also when the words Deep Learning and Machine Learning entered the public lexicon, with the media using these terms interchangeably to explain how AlphaGo competed against human intelligence.
While closely related, these terms do mean different things. And in this day and age, when technology has a large impact on how we live our lives, it’s important to know the difference between these terms.
This article will help you do just that.
The best way of understanding these terms is getting to know the relationship between them. AI, Machine Learning and Deep Learning are like those Russian doll sets - AI is the broadest concept and hence the largest doll in the set, Machine Learning comes in second and Deep Learning can be found within that.
We’ll begin our understanding with the first doll, Artificial Intelligence.
Artificial Intelligence (AI)
Artificial Intelligence as a concept has been a part of our public imagination since centuries, right from Greek myths about mechanical men designed to mimic human behaviour to the more recent Terminator series about super-intelligent cybernetic organisms that travel back in time to destroy/save humanity. It is the theory and development of computer systems able to perform tasks normally requiring human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages.
The concept of AI popularized by Hollywood and Sci-Fi is known as “General AI”- powerful machines that can outperform human beings in all areas.
However, AlphaGo and most other instances of AI in today’s world belong to the “Narrow AI” category wherein these machines are created to replicate or improve human behaviour in one specific task.
The popularity of AI has exploded over the past few years, due with the easy access to parallel computing via GPU’s and the Big Data movement bringing in data of every kind, thus helping explore every depth of possibility in this field.
But how did a field, which was languishing for attention and progress until 2012, make it big so quickly? This question leads us to Machine Learning, the field of Computer Science that has made this development possible.
Machine Learning as a concept implies the ability of a machine to learn for itself, from the data provided. While we usually program machines to follow instructions as per our requirements, in Machine Learning we employ algorithms that systematically parse through the data and learn the behaviour by itself, without being explicitly programmed to do so.
The best example for this would be the classification of emails into different filters, identifying whether an email belongs to Spam or not is done using Machine Learning algorithms.
Machine Learning was conceptualized by the early AI crowd and has since then amassed a lot of techniques such as decision tree learning, naive bayes classifier and support vector machines among others. It has been used extensively in the field of Computer Vision, where the goal is to help computer’s identify different objects within an image.
Initially, this field was riddled with failure - it required a lot of hand coding, processing time and was still unable to match results as per human standards. With time and improvement in technological infrastructure, it’s techniques have becomes extremely powerful but there is one sub-field that has helped advance Machine Learning to the extent that Computer Vision is successfully used by the likes of Facebook to identify objects in a picture and by Uber, Apple and the likes to make driverless cars.
And the technique? That’s what Deep Learning is all about.
Deep Learning is a subfield of Machine Learning that employs the technique of Artificial Neural Networks. It is inspired by human biology - just as our brain consists of a network of neurons firing signals and transmitting information, the algorithm creates a similar setup within a machine with the only difference being that while biological neurons can freely connect with each other, artificial neural networks have discrete layers and connections and follow a predetermined direction.
Essentially, Deep Learning involves feeding the computer system a lot of data which it systematically parses by classifying data via binary true or false questions or extracting a numerical value. This information is stored in the form of neural networks which are then used to classify any form of data - audio, video, speech etc. Though computationally extensive, this technique provides us with excellent result and is now used for a wide range of problems such as navigating driverless cars, re-colouring black and white images, providing medical diagnosis, among others.
In conclusion, it is easy to think of these concepts as concentric circles. AI is the broad goal, the future that is already being realized today. Machine Learning is the most promising approach to making this future a reality. And Deep Learning is Machine Learning’s most powerful technique for making it happen.
For a deeper understand of these concepts, check out our Data Science Courses.