What is Deep Learning ? (An explanation for TechNoobs)
If you spend time online or get updates from the tech industry from time to time, you must have heard about the buzzwords “Deep Learning”, “Machine Learning”, “Artificial Intelligence” and wondered what exactly is that? Is it about the era of AI as shown in movies?
In this article, I aim to decipher its meaning in the simplicity that you can explain it to your grandparents! 😉
TL;DR coming straight to the point.
Deep Learning refers to programs that enables a computer to do something that comes naturally to humans : learn by example. These programs are smart enough to actually “learn” things, without needing someone to explicitly code them. What happens is that the programmer will design a “model”. There is some example data that we give to such a model to show that — “if this is input, this is output”, and the model would learn to map them correctly. After it learns that, we can give it a similar input and it would provide us with an output.
Too hard to understand? Here’s an example to help you. Suppose you have lots of pictures (say a million), and a lot of them contain cats. You want a program to identify which of these pictures contain a cat. What you can do is, you take some of the pictures randomly (say a thousand) and label them as a “cat image” and “non-cat image”. You give one such model as explained above these labeled images. Once this model goes over these pictures a few times, it learns to recognize a cat. You can then run this model against the remaining 990,000 pictures and it would label all of them as a “cat image” or a “non-cat image”.
Do you see its power yet? If you’re interested in knowing more about it — keep reading.
What’s the magic behind it? How is a computer program able to “learn”?
Have you noticed how babies learn to identify daily objects, they learn to speak words and do various activities over time. They observe the environment around them, and they learn from what we teach them. We tell them that this 🍌 is a banana and this 🍎 is an apple and they recognize it over time. Our human neural system is made up of specialized cells called neurons (about 100 billion of them) and these neurons are what enables us to think. When we see an object, our eyes passes the raw image information to the neurons and they recognize what objects that image contains. Similarly, the secret behind Deep Learning is a Neural Network which is made up of neurons (but digital).
A Neural Network basically consists of a bunch of neurons stacked in layers as shown in the figure. The neural network basically tries to fit a complex mathematical function that could map a given input to the output. Each neuron contributes a factor in shaping that function.
The input layer is involved with taking the input and understanding it. The hidden layers add weighted factors to the mathematical function to try to link the input to the output layer where the output is generated. These weights are updated through an algorithm like Gradient Descent to try to get as close to the final function that could give the most probable output.
What’s so cool about Deep Learning?
Well, everything! You have actually seen its applications everywhere, but you didn’t know Deep Learning is making it possible.
- Have you used digital personal assistants like Siri on an Apple device, Alexa by Amazon or Google Assistant by you-know-who on Android phones? These AI bots understand your words (text or voice) and give you an appropriate reply — That’s Deep Learning.
- When you see a few products on Amazon and you see similar products being recommended to you, or when you watch FIFA Highlights on YouTube one day and next day you see all Football/Fifa related videos being recommended to you — That’s Deep Learning.
- Have you heard about self-driving or driver-less cars? Yes, that’s not a joke...they exist. That’s been made possible by guess what — Deep Learning!
- Have you seen face-unlock feature or fingerprint scanner on smartphone devices? Yes — That’s Deep Learning too.
- Have you uploaded a picture on Facebook and you see that the faces are sometimes automatically recognized without you ever telling them? — That’s Deep Learning.
- You obviously have heard about the cool new feature of the iPhone X called “Animoji”. It works because of — Deep Learning.
- Everybody has used Google Translate to translate one language to another — that works on Deep Learning too.
This is just the tip of the iceberg, the list is endless!! New applications and research papers are coming out every week with more cool stuff. Some really cool applications are Image Captioning, Image Generation from sentences, A storyteller that creates stories from an image (see picture below) , Music composition, AI that sees in dark, etc. One article is too short to cover them.
How is Deep Learning related to Artificial Intelligence and Machine Learning?
You must have heard the terms “Artificial Intelligence” and “Machine Learning” more often than “Deep Learning”. Even if you haven’t, here’s a quick definition:
“Artificial Intelligence 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.”
You have probably heard about Artificial Intelligence (AI) more in movies such as “I,Robot”, the “Terminator” series or TV Series such as the “Westworld”. They are mostly depicted as robots who are as smart as humans but are able to process and compute things on a much larger scale and speed.
“Machine learning is an application of artificial intelligence (AI) that provides systems the ability to automatically learn and improve from experience without being explicitly programmed.”
You may have heard this from people and right now, you might be a little confused about the difference between Machine Learning and Deep Learning. It’s all right. Deep Learning is actually a particular kind of Machine Learning which revolves around Neural Networks. Machine Learning is more about algorithms that predict for you, but if there’s an inaccurate prediction, generally an engineer has to step in to resolve it. Deep Learning has a way of checking the accuracy and growing itself.
As clear from the picture, Deep Learning is actually a subset of Machine Learning which in turn is a subset of Artificial Intelligence. The AI right now is made by combining Deep Learning and Machine Learning models at a lower level and adding to them.
I have been into Deep Learning since the past few months and its so cool that I want to tell more people about it. I’m sure many people working on it get asked this question by people with no technical background. So my aim was to explain the idea as simply as possible.
This was my first article on Medium, hope you liked it. I’d appreciate any feedback, feel free to comment or get in touch through email, LinkedIn, Facebook or Whatsapp. \(n_n)/
