Kostiantyn Kasian

Kostiantyn Kasian

Auto-generated profile, based on publicly available information

Kostiantyn Kasian

Zaporizhzhia National Technical University, Zhukovsky Str., 64, 69063, UKRAINE

Research Focus & Metrics

Academic Career Length

11Years (2016-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
16 June 2016 in Journal of information and organizational sciences
Verified
Journal Article
Open access
VSVadym Shkarupylo
Vadym Shkarupylo
ITIgor Tomičić
Igor Tomičić
KKKostiantyn Mykolaiovych Kasian
Kostiantyn Mykolaiovych Kasian
This paper presents the investigation and comparison of TLC model checking method (TLA Checker) properties. There are two different approaches to method usage which are considered. The first one consists of a transition system states attendance by breadth-first search (BFS), and the second one by depth-first search (DFS). The Kripke structure has been chosen as a transition system model. A case study has been conducted, where composite web service usage scenario has been considered. Obtained experimental results are aimed at increasing the effectiveness of TLA+ specifications automated verification.

22 February 2022
Verified
Conference Paper
Closed access
DZDmitry Zhyhulin
Dmitry Zhyhulin
KKKostiantyn Kasian
Kostiantyn Kasian
MKMykola Kasian
Mykola Kasian
This paper describes the proposed method for conducting regression testing of a demo project. The method demonstrates an effective approach to solving regression testing problems by combining change-based prioritization and automating test scripts in whole or in part. The proposed work uses modern automated testing utilities: Selenium WebDriver, Puppeteer and Playwright using the JavaScript programming language. Using automated testing utilities can significantly speed up the time of regression testing. Research methods are based on the development of scripts for the main functionality and end-to-end scripts of the demo project. A comparative analysis of the speed of the automated testing utilities has also been carried out.

1 February 2018
Verified
Conference Paper
Closed access
MKMykola Kasian
Mykola Kasian
KKKostiantyn Kasian
Kostiantyn Kasian
The diagnostic mathematical model of radio-electronic devices that differs taking into account important external factors and their composition is proposed. The diagnostic model includes the mathematical submodel of the device in the node potentials basis, the function of the mean square deviation, which includes measured and calculated values of the output characteristics of the device with weight coefficients, which take into account the accuracy of the measuring devices selected at the stage of ensuring diagnosis. Also, the diagnostic model includes the equation of connection of the selected output characteristics with the potentials in the nodes of the circuit, as well as the equation of the tolerance synthesis for the parameters of the elements, depending on the time and conditions of operation.

1 January 2026 in Scientific papers of Donetsk National Technical University. Series: Informatics, Cybernetics and Computer Science
Verified
Journal Article
Closed access
KKKostiantyn Kasian
Kostiantyn Kasian
MKMykhailo Korpan
Mykhailo Korpan
This work addresses the development of an efficient system for visualizing large point datasets using WebGPU technology and GPU-driven rendering. A three-stage data processing pipeline is proposed that runs entirely on the graphics processor and includes tiling, invisible object culling, and rendering stages using indirect draw calls. The core novelty lies in shifting the entire computational burden from the CPU to the GPU, effectively eliminating the bottlenecks associated with JavaScript execution and data bus synchronization. For the tiling stage, a high-performance compute shader was developed that distributes 10 million points across spatial tiles of 50,000x50,000 conventional units. This process utilizes GPU atomic operations, specifically atomicAdd, to ensure data integrity during simultaneous parallel writes from thousands of execution threads into the tile storage buffers. This approach allows for the creation of a dynamic spatial index directly in the GPU VRAM immediately after the initial data upload. The culling stage is implemented with a separate compute shader that performs frustum culling at the tile level. By employing an Axis-Aligned Bounding Box (AABB) intersection test, the system can instantly discard large clusters of points that fall outside the camera's field of view. This stage generates the parameters for the subsequent rendering command using the drawIndirect mechanism. Consequently, the CPU is never required to read back visibility data from the GPU, which significantly reduces PCIe bus overhead and prevents frame stuttering. A comparative analysis was conducted with a baseline implementation based on dynamic tile streaming from the server at varying numbers of visible tiles from 9 to 400. Experimental studies showed that the proposed approach provides stable performance of over 1400 FPS with a frame time of 0.7-0.8 ms regardless of the number of visible tiles, while the streaming method demonstrates frame time growth from 0.6 ms to 2.2 ms with increasing load. Page load time for the proposed method is stable at 1000 ms versus 100-7,500 ms in the streaming method. The number of network requests is reduced to 20 versus up to 1000 in the baseline method. The experimental testbed included an AMD Ryzen 5 3600 CPU, AMD Radeon RX 6600 GPU, 16GB DDR4-3200 RAM, running Windows 10 Pro 22H2 with the Google Chrome 144 browser. The research confirms that a fully GPU-driven pipeline is the optimal solution for interactive visualization of large static datasets where users frequently change view parameters. The proposed method demonstrates linear scalability with respect to the number of visible tiles (𝑂(1) frame time complexity), minimal network requests regardless of scene complexity, and stable performance independent of network conditions. The streaming method may be more appropriate only for scenarios with extremely large datasets (over 100 million points) that cannot be loaded entirely due to memory constraints, or for users with slow internet connections where gradual loading is preferable.

1 January 2024 in Scientific papers of Donetsk National Technical University. Series: Informatics, Cybernetics and Computer Science
Verified
Journal Article
Closed access
VBVladyslav Butko
Vladyslav Butko
KKKostiantyn Kasian
Kostiantyn Kasian
KAKasian
Kasian
A mathematical formula was chosen, the hardware implementation of which ensures sufficient the result accuracy of the compilers comparison. Based on it, a graph-scheme of the algorithm (GSA) was composed. According to GSA, programs were created in high-level languages (HLL): Matlab and C++. The HLL programs is not adapted to the generation of descriptions in the hardware description language (HDL) with HLS (high level synthesis) optimizations. To confirm the functional equivalence of the HLL programs, they were tested with the HLL test functions. Based on the Matlab and C++ programs, HDL was generated with the HLS tools: HDL Coder and Vitis HLS, respectively. To confirm the functional equivalence of HDL descriptions, they were tested with the HDL test benches. Synthesis and implementation on virtual FPGA (field programmable gate arrays) of generated HDL descriptions was performed. Comparison of the HLS compilers are based on implementation reports and generated HDL descriptions

1 October 2019
Verified
Conference Paper
Closed access
SPSerhii Potapov
Serhii Potapov
KKKostiantyn Kasian
Kostiantyn Kasian
This article discusses a system to accelerate the construction of IoT systems of a smart home, by recognizing interior objects from photographs of the house, with their subsequent transformation into a drawing for manipulating all recognized objects. Various neural network frameworks were considered such as: Apache Spark, TensorFlow, Caffe, as well as methods for recognition: Adam (Adaptive Moment Estimation), methods for optimizing multi-dimensional problems, the backpropagation algorithm. The structure of the system, its components, as well as the operation algorithm are shown. The experiment shows one of the cases with photographs that are taken at an angle and in low light. The experiment demonstrates the program cycle, uploading photos, selecting point of angles, recognizing room objects. As a result, we see the program interface in which a drawing with all recognized objects that we can manipulate, add various elements, such as temperature sensors, various other sensors (lights, etc.) and servos, and with hardware and software for their control.