Have you ever wondered how Google Translates a whole webpage to a special language in a few seconds or your phone gallery groups images supported by their location? All of this is a product of deep learning.
But what exactly is deep learning?
A subset of machine learning, deep learning is also a subset of AI. AI permits machines to mimic human behaviour. Machine learning gives us the technique to realize AI through algorithms. Algorithms train with data and eventually develop a deep learning kind of machine learning inspired by the structure of the human brain in terms of deep learning. This structure is named a ‘man-made’ neural network.
Let us understand deep learning better and the way it is different from machine learning
If I create a machine differentiates between tomatoes and Cherry's using machine learning, we will need to tell the machine the features on which the 2 are often differentiated. These features might be Size and the sort of stem on them. But with deep learning, these differences are picked out by the neural network without human assistance or intervention. However, this sort of Independence is gained at the expense of training the machine with a huge volume of knowledge.
A dive into the working of neural networks
This is best illustrated with an example.
Let assume that there are three students. Each of them writes down a nine-digit on a paper. But the way we write numerals is very varied depending on our handwriting and each one of us had unique handwriting, but these differences are picked by a human bring easily and can read the same number written three different ways by people with 3 different handwriting. Now imagine if a computer had the same kind of knowledge, and this is where deep learning comes in. With deep learning, it is possible to create a neural network trained to spot handwritten digits.
Let us assume on a screen, each number represents a set of pixels. Now that totalled amounts to a much higher number of pixels neurons. At the core of a neural network the knowledge, processing takes place. Each of the pixels gets fed to a neuron within the First layer of our neural network. This forms the input layer on the opposite end. We have the output layer with each neuron representing a digit with a hidden layer existing between them. The knowledge is transferred from one layer to a different over connecting channels. Each of those features a value attached to them and hence is called a ‘weighted Channel’. All neurons have a singular number related to it called bias. This unique bias of each of the neurons gets added to the weighted sum of inputs. Inputs reaching the neuron are then applied to a function referred to as the ‘activation’ function, the results of the activation function determine if the neuron gets activated every time activated neuron passes on information to the subsequent layers.