Deep learning is a technology that is getting a lot of mentions lately. It involves both artificial intelligence and machine learning. It is built on a model that mimics how humans work, building neural networks between machines. As such, deep learning will pave the way for devices and machines to think like humans think.
That may sound creepy to some, but if you think about it, humans now prefer interacting with machines in the same way they interact with other people. Instead of typing to search for something, they simply use voice commands, such as when they ask Siri for location of the nearest Filipino restaurant, or ask Amazon Alexa to turn off the lights. You might be interacting with a machine, but you are doing so in a way that is similar to interacting with another human being.
Interacting this way with machines look simple. But if you look at what happens behind the scenes, you will see complex deep learning technologies at work. It is these algorithms that make it possible for Alexa to turn up the heat at your home, doing it correctly the moment you ask it to.
Deep Learning: Why it is important
Artificial intelligence is well and good, but it will ultimately prove to be useless or frustrating without deep learning. And it is not an easy task to give machines the same means to understand human thinking and concepts. There are a lot of design works involved.
For one, you could structure data in ways that would make it easier for machines and devices to understand. That includes being able to understand the small differences between words by using context. For people, it is easy to use context clues to see what the speaker meant when he or she says “cool” in “that’s cool” and in “this room is cool”, but for machines, you would need to use huge data sets and algorithms to do the same.
How deep learning differs from traditional machine learning
Machine learning is considered to be the precursor of deep learning. What separates machine learning from deep learning is how features are used. For instance, a typical sales transaction would involve several features or information. The item bought, time of sale, item price, any discounts applied, and others.
With machine learning, a programmer would have to design these features, telling the machine what to look for and what a certain feature is. With deep learning, artificial intelligence would figure it all out by itself.
Another example that would better illustrate the difference between machine learning and deep learning is how people are able to tell the difference between a mouse and a dog. Both have four legs, furs, tails, and heads. With machine learning, humans would have a hard time specifying how dogs and mice are different. It would be difficult to specify what characteristics the machine should be looking for, as even humans themselves would find it difficult to name the specific features that make them different.
With deep learning, it is hoped that programmers would not have to make the distinction. Instead, deep learning algorithms would be able to take unstructured data and decide for itself what makes a mouse, a mouse and a dog, and a dog.
Customer experience will be better because of deep learning
Deep learning, therefore, is much better than older machine learning models when it comes to some artificial intelligence characteristics. In fact, deep learning has already proved its mettle when it comes to image recognition and classification, where it shows that it is twice as effective and as accurate than traditional algorithms.
AI models are only useful when it’s around 95% accurate, and arguably, only deep learning algorithms can deliver that.
More than image recognition, you can also apply deep learning to speech recognition to deliver better customer experience. It has made speech recognition technologies very accurate, so much so that speech recognition can now be seen in home automation systems that work with voice command, such as Amazon Alexa.
Some real world examples of how deep learning improves customer experience
The largest horticulture marketplace in the world required their suppliers and sellers to have pictures of the flowers and plants they are selling. These photos needed to be reviewed and checked for accuracy. With the high volume of photos they were getting, having a human validate each photo would have been impossible. With deep learning, however, Royal FloraHolland was able to automate the photo checking process, even if all photos were of plants or flowers, taken from different angles and under a variety of lighting conditions.
University Medical Centrum Groningen (UMCG)
UMCG was able to calculate the volumes of heart ventricles as it changed over time, by using deep learning to go over 4D MRI data. The information UMCG got from this project is used to make decisions about pace makers and other treatments. It involved at least 400 MRI images for every patient taken at different times and at different heart depths. In the past, UMCG would have relied on humans who would be analyzing the MRI scans and doing their own interpretations of these scans.
Deep Learning: What more can you expect?
Deep learning algorithms allow organizations to create modules that are continuously learning. This gives them the chance to develop more complex algorithms that they could use in addition to their existing algorithms, paving way for growth.
It is poised to grow exponentially because of the support of its open source community. The platforms are accessible. Even Apple and other big businesses are opening up their deep learning secrets to the world. This move was a bid to attract better talents to their folds: deep learning and data science professionals would not apply to companies who do not publish their own research openly.
With this change towards open source, there is a lot of hope that deep learning will be able to flourish in the future, and making important contributions to different sectors of society.
Photo courtesy of Ars Electronica.