The world is about to undergo the biggest technological revolution in history with Artificial Intelligence, Machine Learning, Deep Learning, and Computer Vision. However, although there is a lot of talk about these four technologies, the terms are often used interchangeably without any attempt to clearly define their precise meaning.
To best understand how these topics are related to each other, it’s best to take an outside-in look at how they are connected.
Artificial Intelligence is a subset of Computer Science where machines can appear to be intelligent by running programs. It is a very broad topic and covers everything from the automatic doors at the shopping mall to the most intelligent systems built today. From this set of programs, we can refine the set even further.
Machine Learning is the practice of giving a computer a set of rules and tasks, then letting it figure out a way to complete those tasks. The machine in essence starts out with no knowledge, and through trial and error comes up with a suitable solution. The work horse of Machine Learning is the Neural Network.
Neural Networks are algorithms and data structures designed to let machines classify and predict outputs based on a series of inputs. The neural network is an analogous structure to the brain. It consists of Nodes (brain cells), connections, and weights and works on the principle of Gradient Descent. The network has two modes of operation: Training and Inference. In Training mode, lots of data sets are fed into the input nodes and the weights are adjusted. In Inference mode, the unknown data is fed into the input nodes and the system suggests an output. There is a lot more to understanding Neural Networks but this a very broad overview. Neural Networks are usually very complicated and take a lot of computing power to train.
Deep Learning Networks use Neural Networks inside of them. Deep Learning Networks and Neural Networks Architectures have a lot of things in common. They both have an input and output layer and Training and Inference modes. But there are some new twists usually implemented in Deep Learning Networks like Convolution and Max Pooling to make the algorithms run faster and allow for computation at great depths. In a nutshell, one can think of a Deep Learning Network as a network of Neural Networks.
Computer Vision is the practice of giving machines knowledge of their physical surrounding world through sensors. In the past this was a very fragile and complicated task requiring a specific tailored algorithm to analyze pixels. These algorithms were not flexible and had to be used in a specific case and were very susceptible to rotation and lighting. Recent developments in the speed and number of hardware GPUs have allowed Computer Vision to take advantage of Deep Learning Networks that help to mitigate the issues with experienced with standard computer vision algorithms.
The VIA Mobile360 family uses state-of-the-art Deep Learning Networks along with existing Computer Vision techniques on powerful hardware to provide a custom solution for whatever the need is – from harnessing VIA Mobile360 ADAS to alert drivers of possible dangers while driving to leveraging the VIA Smart Facial Recognition System to identify intruders or verify the identity of people entering a building.
About the author:
Jason Lee Gillikin has 6 years experience in the IoT field helping to develop better and smarter products. He holds degrees in Physics and Chemistry with a Masters Degree in Computer Science and Information Systems from the University of North Carolina. Currently he is the Business Development Manager at VIA Technologies helping bring these products to the market. You can contact Jason on LinkedIn or by email.