5. Stand for Wi-Fi Planning Research.

February, 2024

A demonstration example of a neural network model for Wi-Fi coverage in a typical three-room apartment. The model allows: installing up to three routers, visualizing wave propagation, observing wave effects, creating a heat map, fixing acceptable signal levels, and performing an overall analysis of signal coverage quality across the apartment area.

The model was created for a specific demonstration stand with fixed dimensions of 1920x1080 and does not have a adaptive web design.

Click to view

Click to go to the model

4. Additional 50-fold acceleration by GPU

January, 2024

Using graphics processors (GPUs) has become a key factor in the achievements of artificial intelligence and machine learning in recent years. This is because artificial neural networks, like their biological counterparts, are existential systems of parallel computations with a vast number of simple, but simultaneously operating processors, which fully corresponds to the concept of GPU construction. For example, the GPU GeForce RTX 4050 Laptop has 180 times more cores than the CPU Intel Core i7-13700H.

WebGL in 2011 allowed the use of GPU capabilities in browsers, thus revolutionizing the Internet in terms of graphic capabilities. However, WebGL has some fundamental problems. For example, it is quite difficult to perform general-purpose computing (GPGPU), which in our case is the main task.

WebGPU solves these problems by providing an updated general-purpose architecture compatible with modern GPU APIs, full support for graphic rendering, and general-purpose computing GPGPU.

Below is an implementation of a model similar to the previous two, but using a GPU directly in the browser. Considering that the WebGPU technology has only started to be actively implemented in recent years, older or mobile browsers may require an update or the use of Google Chrome Canary. The functionality of the program is limited, but it allows assessing the potential - 70 frames per second for 1,020,272 bodies. For each computational frame, tenfold intermediate iterations are used. As a result, we obtain 700 computational iterations of the second-order differential equations system in a regular browser.

Click to view

Click to go to the model

3. 15 times accelerated wave propagation model

December, 2023

In the second version of the implementation, a much more efficient way of reflecting the magnitude of bodies' deviation from the equilibrium state was found – by writing the state directly into the Canvas matrix:

let canvasData = ctx.getImageData(0, 0, wh, he); canvasData.data[index++] = “color”.

This new approach allowed for a refresh rate on the same platform of 70 frames per second for 270,000 bodies (696 x 389), displayed as 2 x 2 rectangles. That is almost a 15-fold increase. Thus, the program computationally finds the solution to the system of 270,396 second-order differential equations, performing 70 iterations for each equation per second. Experiments have shown that further performance growth is directly limited by the computation of iterations in the equations themselves.

The program has similar functionality to the first implementation, allowing for the formation of medium inhomogeneity and modeling wave processes within it. The offered tool has the ability to build inhomogeneous elastic surfaces, vividly demonstrating all major wave phenomena such as absorption, reflection, refraction, diffraction, interference (principle of superposition), polarization, and dispersion.

The application has intrinsic value both for educational purposes and for research. It should be noted that for such complex media, there are no general analytical solutions similar to the considered wave equation, and the development of processes in them can only be predicted in computational models.

Click to view

Click to go to the model

2. Basic wave propagation model

JavaScript, Canvas, method fillRect()
February, 2023

The first version of the model for wave propagation in an inhomogeneous medium uses a neural network based on the architecture of the cerebellum. Here, tools for forming the medium are presented. In this version, Canvas (HTML5) is used, intended for creating a raster two-dimensional image using JavaScript scripts.

The program operates in two modes:

– field: mode for creating an inhomogeneous medium; provides tools for random filling and shaping with a computer mouse;

– oscillation: mode for direct modeling of wave propagation in the formed medium; oscillators are installed using a computer mouse.

In the first implementation of the model, the display of the state of bodies was carried out through the fillRect() method, providing the output of a filled rectangle at specified coordinates. This approach allowed for a speed of ~5 frames per second for 270,000 bodies in the form of a 2 x 2 rectangle and ~15 frames per second for 68,000 bodies, displayed as 4 x 4 rectangles.

Click to view

Click to go to the model

1. Embodied Cognition

Creating a bio-inspired model that allows predicting and managing complex physical-mechanical processes and their effective interaction with the environment.
December, 2022

The main ideas are outlined in five publications indexed in Scopus:

1. A.R. Nurutdinov, R.K. Latypov. Prospects of a bio-inspired approach in the development of artificial intelligence systems (review of trends). Scientific Notes of Kazan University. Series Physical-Mathematical Sciences. 2022, Vol. 164, No. 2-3, pp. 244-265.

2. Nurutdinov A.R., Latypov R.K. A Novel Artificial Neural Networks Architecture Based on Cerebellum Model. Lobachevskii J Math 44, 733–746 (2023).

3. A. R. Nurutdinov and R. K. Latypov, Cerebellum-Inspired Artificial Neural Networks Architecture, 2023 International Conference on Intelligent and Innovative Technologies in Computing, Electrical and Electronics (IITCEE), Bengaluru, India, 2023, pp. 210-214.

4. A. Nurutdinov and R. Latypov, Bioinspired Neural Network for Simulation of Wave Propagation in Nonhomogeneous Media, 2023 46th International Conference on Telecommunications and Signal Processing (TSP), Prague, Czech Republic, 2023, pp. 230-233.

5. A. Nurutdinov and R. Latypov, Dynamically Predicting Wi-Fi Coverage Mapping Using Bioinspired Neural Networks, 2023 International Conference on Software, Telecommunications and Computer Networks (SoftCOM), Split, Croatia, 2023, pp. 1-6.