## AI - Math. Functions

On this page here is presented a program, which allows to train Artificial Neural Network for creation of model of one of the 3 functions, based on values of argument X and function from it Y. Which is implemented with the help of TensorFlow.JS library.

Among parameters of program there are: function to model; argument values interval, for example [0.0, 1.0]; amount of samples of parameters values and function values for training; amount of layers of Neural Network; amount of neurons in each inner layer. Method, which evaluates errors in functions modelling by parameters values, in comparison with known value. NN parameters optimization algorithm. Samples batch size for training, amount of training iterations, amount of test samples for showing and maximal delta of modelled function value from known one, to present examples as right in the green color.

NNs Learning Complex Functions (TF.JS)

All processes in life are presented as some kind of functions: in the mathematical analytical form or in the form of some kind of a model (precise or approximate). For example, processes in industry can be thought of as the multi-dimensional functions of many arguments. In this case arguments can be technological parameters, equipment states values, price of different technological stages and others. And results can be properties of manufactured component (price, quality and so on) and properties of the industry process itself, like applied energy, time of the process, used resources and so on. This program can be further extended from this simple example to multi-dimensional models of functions of different industry, financial, agriculture and other kinds of processes.

This program is the technical example, showing how Neural Networks can be applied for modelling of different functions. Onwards we'll implement practical tasks, based on TensorFlow library. We'll see how this system works, what is the load, what are the bottlenecks, what can be parallelized and optimized and how this process can be implemented the best in a Binarium network. Then we'll parallelize these calculations in network and will implement tasks for different enterprises, companies, stuios, projects and other kinds of organizations in Russia and other countries. And forth, based on experience of work of this system, we'll implement platform of general purpose computations in Binarium network in all areas.