The Neural Network: Generalized
Ever wondered how we’re able to use algorithms to recognize images, or let alone, identify patterns? Crazy enough, we’re taking a computer and implementing certain human characters within it. Before we initially arrived at solutions for the computer to perform these tasks, we figured that the best way to approach this problem was to model a computer’s thinking process the same way as the human brain. And with this concept, the Neural Network came along.
Our brain utilizes connectivity by sending electrical signals from neuron to neuron through tiny spaces called synapses. These neurons send signals in such an accurate and corresponding manner to one another that they’re able to configure a particular message from the brain that tells the body how to react. Similarly, the Neural Network consists of layers of neurons that each contain a specific “charge”, that when connected with another neuron (through a path of Connectivity Layers), create a path of specific significance that pertains to a certain end result.
Say we were to program an algorithm that allows a computer to recognize a given number image. In order to do this, we’d have to have a set of small unique patterns, that once are all connected, point toward the written number. Also, we must understand that image recognition deals with RGBA pixel values, so our neurons must be given charges (or values) based on the brightness of the particular pixels of the image. To start, say we were told to recognize the number “7” from a given image. Our inputs would first detect a specific pattern from the pixels that corresponds to a certain part of the number “7”. Based on the image below, the input neurons would recognize a pattern that represents a line, which makes up the top portion of the number seven.
After initializing this basic starting point, the neurons utilize the hidden connectivity layers in order to check for other parts of the number 7 that connect to the input. During this process, the “synapse”, or path through which the inputs are connected to the hidden layers and so on, are given a value that’s calculated by taking the product of the neurons. This product is essentially the identifier value for a specific part of the image that the computer is recognizing. The process ends when the input and all layers are connected and point to a specific output, which in our case, would be the number “7”.
So this is it. Although there are many more complications and factors to be taken account of when dealing with image recognition, this is a general understanding of how the Neural Network behaves. For more intricate images, the amount of hidden layers can increase exponentially. After all, we’re trying to model the brain, which is perhaps the most perplexing system in the entire world.
