In recent years, neural network-based differential distinguishers have demonstrated significant advantages in accuracy and effi-ciency over traditional differential distinguishers in symmetric cipher differential analysis. However, when dealing with ciphers involving a higher number of encryption rounds, neural network differential distinguishers still struggle to accurately identify ci-phertext pairs. To address this issue, this study proposes a neural network differential distinguisher model based on attention mechanisms and optimized ciphertext input structures. Specifically, the model first innovates the residual structure within the attention mechanism to maximize the weight of highly discriminative features, enhancing the feature extraction capability of the improved model. Secondly, a multi-scale convolution method is employed, integrating the network structure ideas of RegNet, with the addition of convolutional branches and optimization of activation functions, which further enhances the model's feature ex-traction capability. Finally, a multi-ciphertext input pattern is introduced to improve the input data information, and random key encryption is applied to the input ciphertext structure to construct multi-feature information representations of the ciphertext and encryption functions. The results from 5-8 rounds of experiments on Speck 32/64 indicate that the proposed new neural distinguisher can significantly improve discrimination accuracy to a maximum of 1.65%. On this basis, we carried out an optimization study on the construction method of the multi-ciphertext-pair dataset. The new dataset can increase the accuracy of the distinguisher by 49.16% compared to that of the single-ciphertext-pair case, and can extend the number of attack rounds from 7 to 8.

In today's world, fire forecasting and early detection play a critical role in preventing disasters and minimizing damage to the environment and human settlements. The main goal of the study is the development and testing of machine learning algorithms for automated detection of the initial stages of fires based on the analysis of multispectral images. Within the framework of this study, the capabilities of three popular machine learning methods: extreme gradient boosting, logistic regression, and vanilla convolutional neural network (vanilla CNN), are considered in the task of processing and interpreting multispectral images to predict and detect fires. XGBoost, as a gradient-boosted decision tree algorithm, provides high processing speed and accuracy, logistic regression stands out for its simplicity and interpretability, while vanilla CNN uses the power of deep learning to analyze spatial and spectral data. The results of the study show that the integration of these methods into monitoring systems can significantly improve the efficiency of early fire detection, as well as help in predicting potential fires.

The number of new cybersecurity threats and opportunities is increasing over time, as well as the amount of information that is generated, processed, stored and transmitted using ICTs. Particularly sensitive are the objects of critical infrastructure of the state, which include the mining industry, transport, telecommunications, the banking system, etc. From these positions, the development of systems for detecting attacks and identifying intruders (including the critical infrastructure of the state) is an important and relevant scientific task, which determined the tasks of this article. The paper identifies the main factors influencing the choice of the most effective method for calculating the importance coefficients to increase the objectivity and simplicity of expert assessment of security events in cyberspace. Also, a methodology for conducting an experimental study was developed, in which the goals and objectives of the experiment, input and output parameters, the hypothesis and research criteria, the sufficiency of experimental objects and the sequence of necessary actions were determined. The conducted experimental study confirmed the adequacy of the models proposed in the work, as well as the ability of the method and system created on their basis to detect targeted attacks and identify intruders in cyberspace at an early stage, which is not included in the functionality of modern intrusion detection and prevention systems.

This study presents an automatic system for baking bread in production. The article deals with the tasks of automated control of the bread baking process. A convection oven is considered as an example. The goal is to use methods to control the stage of baking bread, temperature and humidity using a fuzzy logic algorithm. System processes first receive a value compared with the set ones, after that the adder reports the difference between the humidity and temperature set by the sensor, and transmits the data to the fuzzy logic blog. And fuzzy logic makes a decision in accordance with the base logic. This article discusses various methods and systems for automatic process control in the food industry, and predicts the future of automatic process control. All input parameters were controlled simultaneously using a fuzzy control system developed during the study. The fuzzy control system was created using MATLAB Fuzzy Logic Toolbox, and the operating structure of the control system was analyzed. An intelligent control scheme based on a fuzzy controller is proposed. All input data obtained during bread baking are compared with expert data, and rule-based solutions are selected depending on the results obtained. Optimization of temperature and humidity in the chamber ensures high quality of the baked product. As a result, the circulating air is heated and mixed, ensuring the uniformity of baking bread. With the help of convection, hot air flows are distributed throughout the chamber, ensuring a uniform temperature in all parts of the oven. Bread acquires volume and softness, becomes ruddy on all sides, and after intense thermal exposure, glossiness appears in the crust of bread.

The present study provides a systematic benchmarking of nine state-of-the-art object detection models adapted to a specialized animal dataset. The main objective was to evaluate their performance in terms of accuracy and inference speed in the context of real-time and commercial applications within the agro-industrial sector. The method-ology involved aggregating and thoroughly cleaning several cattle image datasets, fol-lowed by fine-tuning models with diverse architectures — two-stage, one-stage, and transformer-based. Performance evaluation was conducted on a unified hardware platform equipped with an NVIDIA RTX 4090 GPU with 24 GB of VRAM, which is critical for the objec-tive assessment of speed-related metrics. Key results show that the D-FINE and Co-DETR models demonstrated the highest ac-curacy (AP@[IoU=0.50:0.95] of 0.872 and 0.851, respectively), while RTMDet and YOLOv11 proved to be the fastest (15.81 ms/image and 19.14 ms/image in reproducible mode). A general trend was observed toward significant improvement in average pre-cision (AP) metrics on the specialized dataset compared to their known results on gen-eral-purpose datasets such as COCO, while relative inference speed rankings remained largely consistent.