No. 22/2025

Online archive of Computer Science and Mathematical Modelling

No. 22/2025

1. Maja Czyżewska - Deep learning architectures for financial time-series forecasting: A comparative methodological review

   Pages: 5 - 12
   Abstract: This paper presents a comparative review of selected deep learning architectures used in financial time-series forecasting, with particular emphasis on Long Short-Term Memory (LSTM) networks, Convolutional Neural Networks (CNNs), and attention-based hybrid models. The discussion is framed around three representative market environments: cryptocurrency, equity index, and foreign exchange data, in order to highlight how differences in volatility, liquidity, and market structure may affect model suitability. The paper outlines the main challenges of financial forecasting, including nonlinear dynamics, temporal heterogeneity, and sensitivity to changing market conditions. Particular attention is given to hybrid and attention-based approaches, especially the Regularized Self-Attention Regression (RSAR) model, as an example of a more adaptive forecasting framework. The review suggests that no single architecture should be regarded as universally optimal for all financial series. Instead, the suitability of a given model depends on the statistical properties of the analyzed data, the forecasting horizon, and practical constraints such as computational complexity and sensitivity to hyperparameter selection.
   Keywords financial forecasting, deep learning, hybrid models
   Full article:
2. Bartłomiej Firlej, Przemysław Sujecki - Autonomous decision system for UAV control in 3D simulated environment using deep neural networks

   Pages: 13 - 20
   Abstract: This article presents an autonomous decision-making system for unmanned aerial vehicles (UAVs) operating in GPS-denied environments. The proposed solution is based on reinforcement learning and is intended to support mission execution when external positioning signals and continuous operator guidance are unavailable or unreliable. The system is implemented in a Unity-based 3D simulation environment and employs a PPO-trained Actor–Critic neural architecture with continuous control outputs. The primary operational scenario considered in this work is the autonomous protection of a mothership drone, where the UAV must maintain effective defensive positioning, preserve mission continuity, and react to dynamic threats and environmental constraints. In addition, a strike mission is included as a secondary task configuration to demonstrate the adaptability of the proposed framework to different UAV operational roles. The article discusses the conceptual structure of the system, the observation design, the control formulation, and the role of reward engineering in shaping autonomous mission behavior.
   Keywords deep neural networks, UAV, autonomous
   Full article:
3. Jakub Grątkiewicz - Modeling of the discrete RGB digital color space to the visible color spectrum

   Pages: 21 - 30
   Abstract: The author asks a basic question: how does the world of color work, and how does a continuous physical spectrum become a perceived hue. This perspective is then contrasted with digital practice, where colors are encoded discretely as RGB triplets and reproduced by displays that emit artificial light with device-specific spectral characteristics. The article examines whether a conversion from digital colors to a physical, wavelength-based description is possible, and whether such a conversion is meaningful. By considering metamerism, inter-device variability, viewing conditions, and observer-dependent perception, the author argues that assigning an RGB value to a single "true spectrum" is inherently non-unique and only weakly justified from a perceptual standpoint. A further conclusion is practical but important: even the finite RGB space is beyond direct human comprehension. Not all differences between RGB codes are perceptually discriminable, and even when they are, discrimination does not imply that people can reliably assign unique names to every visible difference. In effect, the infinite spectral domain is reduced to a finite set of digital RGB codes and then reduced again to an even smaller set of linguistic color categories. Building on this, the paper proposes heuristics that are sufficient for precise everyday use of color terminology, while acknowledging their heuristic nature: (1) an author-defined RGB color cube with vector-based region assignment for named colors, (2) quantitative sampling with a nearest labeled neighbor rule, and (3) data-driven classification on the sampled datasets, including supervised models and clustering methods.
   Keywords computer graphics, colors, ML
   Full article:
4. Mateusz Mohr - .NET 10 Common Intermediate Language

   Pages: 31 - 36
   Abstract: The .NET Common Intermediate Language (CIL, IL – Intermediate Language) is a programming language that is an intermediate version between source code and machine code. It is standardized and closely linked to the .NET platform and high-level languages such as C#, F#, and Visual Basic .NET (VB.NET). The .NET platform uses a two-step compilation process. It converts high-level source code (e.g., C#) into a .NET platform assembly containing CIL code in the form of an executable file (.exe) or dynamic link library (.dll). Only at runtime is this intermediate code, loaded by the .NET CLR (Common Language Runtime), translated into native processor instructions appropriate for the specific operating system and hardware architecture. Thanks to this approach, the same CIL code can be run on different platforms such as Windows, Linux, or macOS without having to recompile the source code. Although intermediate language code is not written directly by the programmer, it can be analyzed and edited using special tools such as ILDASM and ILASM developed by Microsoft, or ILSpy, a tool developed independently of Microsoft as an open-source project. The characteristic two-step compilation process makes it relatively easy to analyze the source code of a program compiled and delivered as an executable file, and even to recreate it in a form similar to the original code.
   Keywords .NET, CIL, Common Intermediate Language
   Full article: