Have you ever wondered how your phone sorts out spam emails, how your keyboard predicts what you are about to type or how cars are able self-drive? All of these new and upcoming technologies are the result of a very famous term, we use it every day, we see it everywhere, it is “Artificial Intelligence”. Often the term A.I is spoken of when we talk about a distant future, but that future is now, we are already at the second wave of artificial intelligence and the era of true artificial intelligence is upon us.
I am sure a lot of us have heard artificial intelligence and most people want to ride the wave that comes along with it. Almost everyone wants to get a job in the upcoming industry as it shows promise for all fields. The fourth wave of artificial intelligence -- true intelligence and human like responses, also known as sentience – is approaching, it is only a matter of time but like everyone else we must take the first steps. The first steps towards artificial intelligence is Machine or Deep Learning, a method through which computers can learn and return results based on their understanding. For a computer to formulate its own answer without specific instruction to do so, that is what the current level of learning is for a machine.
A neural network is made of nodes -- similar to how we have neurons -- that apply transformations to the data. As most systems its basic operation is: it is given an input and generates an output. However unlike most other systems we do not give them the information on how to generate the answer. We instead set them up so they can learn from massive amounts of data that already have the input and generated output. This data is known as “training” data. However, it does not just read the data and simply construct its understanding of the data, it instead modifies its previous understanding of the data by running its data through the series of layers in the model of the neural network and compares the generated output to the actual output. Based on the differences in the outputs, it modifies the weight in the node and once more takes a piece of training data and tests again. Therefore, through a process of trial and error of over 1000’s of times, the neural network will be able to learn.
The neural network consists of a series of layers, typically an input layer, at least one hidden layer and finally an output layer. If the network has more than one hidden layer we often to refer to it as “Deep Learning” whereas a network with just one hidden layer is called “Machine Learning”. The input layer is straightforward as it is a layer of nodes that take the input that must be run through the network. The structure of the network is similar to a mesh topology with every node in a layer is connected to every node in the layers next to it. The input layer nodes then pass their data to the hidden layer, and this is where it gets interesting. The hidden layer, also consisting of a collection of nodes, has weights and biases that modify the data passed through it using different operators. By modifying the data we will end up with different results in the output. Usually we start by taking random weights and nodes and they will change as the network processes data to try to achieve the actual output. Finally the output layer which can holds the outputs after the input data has been modified by the weights. In the case of a classification neural network -- a neural network that tries to classify data into different labels like a movie review being positive or negative -- the neural network will have a set number of outputs based on the label you are trying to classify it for. The other type is a prediction neural network it will take the input and give a single output which will hold the prediction for the given input data.
Copyright © 2018 Sahas Gelli - All Rights Reserved.