Gentle and basic steps to Neural Network.
Neural networks are everywhere and chances are high that you are using them every day. When you're shopping online and a product you like pops up, when you see a weather forecast or stock market and why not when you use your smartphone? Although there exist popular and professional tools such as PyTorch or TensorFlow that let us train neural networks from a high-level perspective. But it is very important as a person who starts its journey into artificial intelligence and machine learning to brush up on the basics and understand the details of the algorithm behind the neural network. Therefore, this will allow us to fully understand what's going on inside networks.
A neural network is a piece of software. It's a model of the brain, and like so, it's capable of reproducing some behaviors of the brain like learning and classifying. So it is important to understand the crucial elements in the human brain. As a very basic example telling the difference between apples and oranges in a picture.
Now let's consider what a neural network is not so that we get some misconceptions out of the way. A neural network is not a series of if-then statements or any other form of hard-coded data the software may fetch to produce an answer. A neural network is not all there is to machine learning. In fact, you'll see that machine learning is an extensive field of knowledge on its own. And lastly, although neural networks are beneficial for classifying. Now let's understand in an easier way: a neural network is a collection of neurons, which are small computing units, that perform simple calculations to collectively solve a problem. These neurons may be interconnected in several ways. Let us consider among the dozens of neural network like recurrence and so on just three very special types:
- Hopfield Neural Network
- Feedforward Neural Network
- Convolutional Neural Networks
Hopfield neural networks, which have a fully connected architecture. That is every neuron in the network sends its output to all the other neurons. It does have inputs and outputs. The inputs modify what's going on inside the network, including the output values. Perhaps the best-known category of neural networks is the feedforward model, where we have a set of inputs, a series of layers of neurons with signals propagating forward until they reach the output. The success of this type of neural network led to the development of a large family called deep neural networks, which have a large number of neuron layers between the input and the output. One example of this type is the evolution of deep neural networks are convolutional neural networks.
Neural networks were inspired by the brain, so it makes sense to take a quick view of the biological brain. The brain is made of neurons. The brain is a biological neural network. Each neuron in the brain performs a cognitive function. Finally, the combination of these cognitive functions results in our brain's mental ability. Everything the brain does is the result of these tiny computations like feelings, memories, movement, speech, dreams, all of it. The structure of biological neurons consists of various parts as followed. Dendrites take electrical signals coming from nerves or other neurons. These signals may come from sense organs like the eyes, tongue, or ears. The nucleus of the neuron is where some processing is believed to happen so that the combination of inputs can yield some binary response so that if the stimulus that obtaining from the input is high enough, it is said that the neuron fires an output signal. This output signal goes through an extension called the axon into other neurons therefore scientists build a computational model of a neuron based on these three important elements.
