Have you ever imagined a world where computers can think for themselves like humans? For those of you that saw the movie ‘Her’, you will understand what I mean. In this movie, the lead character falls in love with his computer OP (Operating System), Samantha. When we think of an AI (Artificial Intelligence) OS that can think and make decisions like Samantha, we imagine a futuristic AI computing system that combines ‘situational recognition’ capabilities and ‘Deep Learning’ technology. Deep Learning has received a particular amount of spotlight as a next generation industrial IT trend in 2014. Today we will take a detailed look at the Deep Learning technology that goes into AI.
Deep Learning, the Shortcut to Thinking Computers
Deep Learning uses 2 methods to classify any objects. These methods are supervised learning and unsupervised learning. When a computer uses clustering and classification to recognize, infer and make judgment on objects and data without a standard of judgment, it is known as unsupervised learning. Conversely, when a lot of data is entered beforehand so that a computer can make judgments, it is known as supervised learning.
Deep Learning is the data analysis capabilities of human intelligence and brain functionality being used to make predictions. Like a human detecting patterns in a large amount of data, a computer distinguishes objects. Also, the basic idea of a computer is structured like the neural circuit of the brain made up of neurons and synapses. So we can expect results from a computer to be like that of a human analyzing systemized data. Since a computer system is structured like the neural circuit of the human brain, it follows that a computer could posses the creative capabilities of a human. If we look at automobile recognition systems, Deep Learning AI is able to determine the owner of a vehicle by detecting the color, license plate number and model of the car without entering the information of the person.
Deep Learning uses a data management structure format known as the deep neural network algorithm. Since the deep neural network algorithm is structured to have one neuron connected to many neurons, it is possible to manage signal dispersion. Also, the neuron movement is non-linear, which make management of not structured data possible.
The Rise of Deep Learning
Deep Learning is receiving a lot of attention now but Deep Learning research began back in the 1980s. At that time, the slow speed of attaining a high level of accuracy in neural network learning was an obstacle. Furthermore, the inability to resolve this overload of data caused people to think there were no practical applications for the technology. Below are 3 reasons for the recent sudden re-emergence of Deep Learning.
The first reason is that a weakness in artificial neural networks was overcome. In 2004, a new Deep Learning algorithm known as RBM (Restricted Boltzmann Machine) emerged on the scene and introduced the idea of ‘Drop-out’, which makes it possible to arrange data simply and prevent overloading systems.
Hardware is at the core of the Deep Learning revival. The powerful GPU (Graphics Processing Unit) has demonstrated the ability to reduce work times for calculations combining complex matrices and vectors from several weeks to just a few days. IBM is also creating an artificial intelligence ecosystem with their internally developed ‘Watson’ super computer.
The final reason for the revival of Deep Learning is the advent of the big data era. The large amount of information pouring in from online is processed and reproduced with high-volume data analysis capabilities and this is a large part of learning. Deep Learning meshes big data together and that data can be used in many different fields. Among those fields is sound and image recognition. Companies from around the world are battling to get a foothold in this industry.
Deep Learning has become a major trend in the changing technological environment. I feel that the AI we’ve been seeing in only in movies for so long is actually not that far away. In part 2 of this series, we will discuss how AI is not only a popular craze but also a true possibility and look at how Korean companies are fairing in the field.
Written by Jongho Choi, University reporter for LG CNS