This project is a major overhaul of the initial pattern generator presented in the original project by @HDKnatch, with a larger, faster neural network. This version of the pattern generator has a total of 53 nodes. an input layer of 10 nodes, 4 hidden layers of 10 nodes each, and an output layer of 3 nodes. It can support up to 4,096 characters, and works by precomputing most of the weighted sum, divided by a value that stops the scaling of the result from going out of control, and then uses the precomputed values, plus the coordinates value to calculate the output. You can press space to give the AI a new weight set. It uses No activation function for the input layer, the Tanh activation function for the hidden layers, and the relu activation function for the output layer. It is recommended that you use turbowarp to run this project; https://turbowarp.org/1052908835?fps=250&turbo
A neural network is a mathematical model that is capable of learning and memorizing patterns in data. They are comprised of "Neurons", which are mathematical nodes that work roughly based on our understanding of how real biological neurons work. The neuron takes the sum of all values in the previous layer multiplied by their respective weights, adds a bias, and puts the final result through an activation function, which is a mathematical formula that creates a non-linear output, allowing the neural network to therefore provide non-linear outputs. This neural network uses a random weight set, and takes in a total of 10 initial values to transform into a complex output. The first 8 values are determined by a simple algorithm that takes in the maximum of 2048 characters that can be given to it at once, and puts them through the sine function, multiplies them by a large number, modulates them by 1, and multiplies them again by 16. The final 2 input values are simply the x and y positions scaled down to a range between 16 and -16. It then provides three output values, which are scaled to an integer (rounded) between 0 and 255. These values represent RGB color values.
