AI vs Machine Learning vs Deep Learning
If one was to go through the technology roadmap of any organization worth its salt, amongst other things, the presence of these two alphabets — ‘AI’, is almost a certainty. The other phrases associated with AI ( artificial intelligence) you are most likely to find are Machine Learning and Deep Learning.
In some cases these phrases get used interchangebly — in most cases, not entirely accurately.
So here is a quick post to help explain some of the terms a little better.
I am your Superset, Luke
Although the title of the post says AI vs Machine learning vs Deep Learning, actually it should read AI ⊃ Machine Learning ⊃ Deep Learning, because these are not distinct fields but rather subsets of each other. The figure below explains the relationship.
Artificial Intelligence systems, as they started off and in their simplest forms were rule based and knowledge based systems. They worked fine in the scenarios, where the ‘knowledge’ was codified and the system could be representated by a finite set of rules. For instance a game of chess or ‘if she says ‘Hello’, then say ‘Hi’ kind of scenarios would be good examples.
But of course the world is not codified, and so these systems were limiting. Majority of the knowledge is tacit and informal; and so impossible for us to write down in the form of rules and instructions for a computer to follow. Enter Machine Learning.
With Machine Learning algorithms, based on prior data points, the system identifies patterns and is able to arrive at supposedly subjective decisions. When the data is available on the right set of features — ‘Is this mail a spam or not’ ( classificiation) or for that matter ‘how many comments is this article likely to garner’ (linear regression), would be some examples. Oh and by the way since we are supervising the system by feeding it previous data points , with right answers ‘labeled’ as such, for it to learn from, its called Supervised Learning ( another term you are likely to hear a lot).
Where the system on its own groups or ‘clusters’ data points based on similarities, without any supervision from us, it is called — Unsupervised learning. Any kind of clustering exercise would be a good example — identifying a particular segment of customers expected to respond to a product in a certain way, or identifying a certain type of tissue in a given lab culture etc.
But the problem is — identifying all the features that are needed to make the model an accurate representation of the scenario being processed.
The solution plausibly is — that the system not only discovers the output based on the representation, but learns the representation itself. This is known as Representation Learning.
But every representation will have various factors of influence causing variations, and for the system to learn the representation we need to identify which factors to discard and which to keep. This is of course a very difficult task when it comes to complex represenations.
So the solution, that has gained immense popularity since late 2006, and largely responsible for the revival of interest in Artificial Intelligence ( yes, its an old field, dates back to the 1940s, I believe. But more on that later), is representing complex representations in terms of simpler presentations — Deep Learning. This is done through muliple layers of abstract factors with each adding to and providing a new representation.
Google Artificial Brain recognizing the video of a Cat in 2012, is probably one of the earliest and most talked about breakthroughs in Deep Learning. The way Deep Learning works for instance is given picture of a cat — it starts by identifying the edge of the face > identifies featres like the eye, nose, whiskers & so on and thus arrives at a conclusion that it must be some sort of an animal.
Hopefully this post gave a little more insight into the terms. Do leave your comments & thoughts — including any better ways to explain the terms. I will be happy to incorporate them.
Disclaimers: The above post in no way claims any copyright to any of the images ( cluster, cat and banner image) or literature presented.