On this page the program is presented, which allows to train neural network to learn model of a function, chosen by user, which is based on TensorFlow.JS library. Analitical form of function is used only for training of a network and for checking of correctness of training. Chose a function, define an interval of parameter X, amount of function's samples and parameters of neural network and of a training and press Train Model batton. Further on we will develop this program for learning of "functions" of industry, financial, agriculture and other kinds of processes for subsequent analysis of such "functions", for example, for searching of their minimums and maximums.
https://binarium.money/ai/test1https://binarium.money/info/aiNeural Networks are models of complex multi-dimensional functions, which often can not be presented in analytical form. While learning, they build model of such function, based on its input parameters and according result; then there is ability to work with such models. For, example : functions of manufacturing processes, functions of equipment state, functions of ideal AI bot behaviour in games and others. Then, based on these models, there is ability to get appriximate results of functions for parameters values, which were not present in training input parameters set. For example, there is ability to find minimums and maximums of a functions on defined parameters intervals : minimize energy cost, time of production or maximize manufactured equipment quality, find the most optimal factory eqipment usage mode, or find the most optimal AI bot behaviour in a game in current surrounding environment. There is ability to analyze such functions in hand mode or there is ability to automate such processes with the help, for example, of Genetical Algorithms for searching of global maximums and minimums of functions on intervals or findings of points with other interesting properties. Speaking simply : Neural Networks - are very advanced interpolators of input parameters, which allow to obtain results, which are close to values of modelled functions and which can learn and remember their features :-) .
This program can be easily adapted to multi-dimensional case, in which parameters can present any industrial (temperature and duration of a stage of manufacturing process, time of produced compnents cooling, ventillation intensity and others), financial (daily trading volume, oil prices, foreign currencies exchange rates, and others), agriculture (watering duration, application of fertilizers and their amount, and others) and other kinds of data. Results also can be any real data: applied energy, time and resources for products creation, products cost, amount of a harvest and others. This data can be taken from, for examples, statistical data of enterprises and quality control services or can be measured in experiments.
Among parameters of a program there are : selector of function to be modelled; selection of X argument interval of a function defenition, for example [0.0, 1.0]; amount of samples of argument values and function results from them for training; amount of layers of a neural network; amount of neurons in each internal layer. Function, which estimates errors in NN modelling of a function, based on parameters values, in comparison with known values. Network parameters optimization algorithm for training. Learning batch size, amount of iterations, amount of examples for showing and maximal delta of modelled value from known real one, for showing of examples as correct in a green color.
This program - is a technical example of how Artificial Neural Networks can be used for modelling of different functions. Further on we will implement practical tasks, bases on TensorFlow library. We'll look how this system works, what is the load, what are the bottlenecks, what can be parallelized and optimized and how this all can be implemented in Binarium network. Then we'll parallelize calculations for AI in Binarium network and will implement tasks for different enterprises, companies, studios, projects and other organizations in Russia and other countries. Then, based on experience of work of this system, we'll implement platform for general purpose computations performing in all areas.
If cryptography technologies and Block-Chain ( or Directed Acyclic Graphs ) technologies will be combined for provisionning of honesty and justice and Artificial Intelligence for optimization of different processes, then we obtain new model of economics, which did not existed before and which has much lower operation cost. Because there is no need in big staff and there is no need in paper documents workflow, there is no need in big amount of government institutes : controlling ones (part of functions of institutions, which perform economical control) and providing legitimacy and honesty of economical operations (for example, banks). Which operates much faster, because everything is in electronic form and trained AI reacts on different events much faster and more precisely, than humans. Also, taking into account experience of countries, which actively implement Artificial Intelligence techniques, optimization of different processes with use of AI increases their efficiency by 5 - 15% in comparison with their organization, based on experience of people and work of expert systems. Which, for example, with monthly operating volume 1 billions of Rubles gives economy in 50 - 150 millions of Rubles. If we will combine here manufacturing automation and robotic technologies + nano-technologies + details 3D-printing + laser processing + Solar energy or thermo-nuclear reactors + medical and biological technologies + technologies, which combine different areas, for example NBICS-technologies, then humanity enters new era of its existence. It can be era of unseen prosperity and flourishing of all humanity or era of unseen slavery of most of the humanity under small group of people, which control technologies and understand how they work and which stupefy the others to keep them under control. But, maybe, everythin will be in a blurred and partly implemented state - everything depends on people.
