In the tech industry, terms like “Machine Learning”, “AI”, and “Deep Learning” are thrown around very casually, often without much context.
Let’s take one of these terms, “Deep Learning”, and dive into what it really means, and why you should care.
At the most fundamental level, a computer is an instruction-following machine. You give it a sequence of steps you want it to carry out, and it carries out those steps as quickly as it can.
This works great when you want to solve a clearly defined, deterministic problem, like sorting numbers or serving up web pages. But what happens when you want to solve a problem like categorizing photographs based on their content?
Imagine that you have a set of images, and you want to sort them into two sets: images which contain a bird, and images which do not. How would you write the steps to do this? There is no clear, defined set of steps you can program into the computer which can accurately differentiate “bird” and “non-bird” images. Instead, we need to give the computer a way to learn on its own, perhaps using a pre-tagged set of “bird” and “non-bird” images.
One approach to this problem is to use artificial neural networks. These are computer models that approximate the neural networks that exist inside the human brain. Artificial neural networks are much simpler than their biological counterparts, but retain some of the same basic properties.
So what is “deep learning”? Deep learning is a machine learning technique that uses artificial neural networks consisting of multiple layers. The “deep” in deep learning comes from the fact that the artificial neural network is itself “deep” (has many layers) and is not meant to imply that deep learning is somehow more insightful, more powerful, or more complete than other machine learning
So why has deep learning suddenly become so popular? Is it better than other machine learning techniques? This is a hard question to answer without getting very technical. For AI and machine learning experts, deep learning is just one tool in the toolbox, albeit a very flexible and powerful one. The reason for the sudden popularity of deep learning has a little bit to do with improvements in how neural networks are trained, and a lot to do with increases in computing power, especially GPU computing. Before GPU computing became more affordable and widely available, large neural networks of the type used in deep learning took too long (or were too expensive) to train.
Now that deep learning can be more easily and cheaply implemented, we can expect to see more and more successful applications of this powerful learning tool. Already, deep learning is being used successfully in fields such as speech recognition, image recognition, natural language processing (including translation), fraud detection, and more.
But don’t forget! Deep learning is just one technique. See these articles for more information:
- The difference between deep learning and machine learning
- What is supervised learning?
- What is unsupervised learning?
- The difference between supervised learning and unsupervised learning