Whether it is autonomous vehicle, real time translation or A.I. beating world’s best player of Go, Deep learning is at the heart of all these technologies. Due to the advancement in the computational ability we are now able to implement deep learning algorithms to solve various complex problems. For example, “Asked whether two unfamiliar photos of faces show the same person, a human being will get it right 97.53 percent of the time. New software developed by researchers at Facebook can score 97.25 percent on the same challenge, regardless of variations in lighting or whether the person in the picture is directly facing the camera.” This is now all possible because of complex model based on deep learning.
Deep Learning is a subset of machine learning which is build on the concept of how our cerebral cortex works and took inspiration from neuroscience. The basic functionality of neuron can be represented mathematically by a perceptron. As shown below, here x,y,z are inputs w0,w1,w2 are weights (how important is a particular input) and after summation and an activation function (which type of values -ve or +ve or both values to consider) we have the output. There are several types of activation functions which are out of scope of this article. Please refer here for further information. By combining several of these perceptrons we create new layers and those layers together create an artificial neural network (ANN).
In a typical neural network, we have at least one input layer, one hidden layer and one output layer. A typical neural network shown below can have few hidden layers where each hidden layer has a specific function either to process the input from previous layer or to extract specific information.
Now to understand how deep learning is used in various area let us consider an example of classification in computer vision. When we see a dog or a cat, we usually identify them with certain attributes and features. When we take several images of cats and dogs and feed them into a neural network the network will be able to extract those features itself. Once trained, when we will feed a new image of a cat or a dog that network will be able to identify the subject correctly as shown below.
Classification problem can also be solved by supervised learning which is explained in this article. The difference between both cases is shown in the picture below. In supervised learning we must extract the features and provide the right labels on the other hand in deep learning the network can extract the features itself.
To learn more about the solution using deep learning to solve real world use-cases visit Tech Data’s IoT solutions catalogue page here.
IoT Solution Architect (A.I and Vision)
IoT and Analytics Team
Tech Data Europe