Just when you thought you’d got your head around the whole Machine Learning thing…BAMN! There’s a new tech buzzword in town rearing up to take it’s place.
And while it may seem like just another Silicon Valley buzzword that all the new startups will claim to be using, deep learning is actually already being used to make some really astounding advances. We’re talking borderline science fiction here.
We caught up with deep learning expert, Andrew Ng, and asked him to explain what deep learning is and how we should expect to see it change the world in 2016.
What is deep learning?
Deep learning is a subset of machine learning that essentially refers to trying to map neural networks (the same stuff that makes your brain work). By mapping neural networks we can recreate some of the same processes that the human brain performs.
The goal is to create algorithms that can take in very unstructured data, like images, audio waves or text blocks (things traditionally very hard for computers to process) and predict the properties of those inputs. The classic example is getting computers to recognize images of cats.
That may not sound exciting if you’re not a computer science wiz, but actually, dealing with real world problems – like cat detection – requires some very complex functions. And once we work out how to run those complex functions, we can start to do some pretty stellar stuff.
Why is deep learning a big deal now?
Neural networks are nothing new. We’ve been trying to map neurons in the brain since the 1890’s and using those principles in computer science since the late 60’s.
What’s changed, Andrew says, is the scale at which we can do these things. Increased computing power has allowed us to map and process much larger neural networks than ever before. We also have a lot more data that we can use to train these networks.
Believe it or not the way you train a computer to recognize a cat is by showing it thousands of cat pictures.
Astonishing things deep learning can do
So what does all that added data, computer power and neural understanding really mean? How is any of this actually being used in the real world?
Computers writing captions
With the use of neural networks, computers can not only recognize pictures of cats they can actually describe what those cats are doing. Think for just a second about how much more brain power that takes than simply saying “yes there is a cat” or “no there is not a cat”. A lot.
Taking that brain power to the next level, you can even ask questions of a computer about the image you showed it. So it could tell you that the bus is red or the cat is on the floor.
Seeing for the blind
If computers not only recognize images, but understand and interpret them the same way our brains can, then we can effectively mimic sight for people who don’t have any.
The people at Baidu have created Baidu Light, a wearable device that can take pictures of things and return the caption to the wearer. They could take a picture of their money at the register to find out how much they have and in what denominations, take pictures of the contents of their refrigerator and so on.
Improved speech recognition
Neural networks aren’t limited to visual inputs. Scientists and researchers have also been using their brain power to improve speech recognition. With deep learning Andrew’s team at Baidu have managed to improve speech recognition from 89% accuracy to 99% accuracy.
That may not sound like a lot, but when you consider that with 95% 1 in every 20 words would likely be wrong, 99% is game changing. This is especially important in Baidu’s home country of China where much of the population is illiterate. Without good speech recognition, many of them wouldn’t have access to the internet.
Andrew sees the future of speech technology going far beyond your cellphone. One day soon, he says, you’ll be talking to your car, your home appliances and your wearable devices.
Predicting user behavior
With more and more user data being fed into their algorithms, the team at Baidu have begun to use deep learning to predict both user and machine behavior. They know which ads you’re most likely to click on, which servers are most likely to need repair and where security threats are most likely to come from. All of these advancements have had a direct impact on their revenue.
Deep learning in your pocket
These examples are certainly impressive, but many of them aren’t particularly tangible – especially if you don’t live in China where Baidu is launching these services.
If you want to actually get your hands on something built with deep learning you can download Faceyou (available on the Apple App Store). It’s a mobile app that can detect your face in real time and either add props or swap faces with another image.
This technology is only possible because of deep learning, which allows them to map your facial features and accurately apply the elements that appear to move with you. Right now it’s just a fun app, but imagine how the retail space would change if you could see yourself in clothes or with a hairstyle online without having to go and try anything on.
Giving computers superpowers
As Andrew explains it, deep learning is effectively a computer’s version of the radioactive spider from Spider Man. If computers can be taught to see, hear and understand like humans it will become a lot easier to interact with them.
The progress of deep learning is growing by leaps and bounds. 2016 is surely going to be a very exciting year in the deep learning space.