The increase in threats to network infrastructure is closely linked to the need for developing new, advanced, and efficient methods for protecting IT infrastructure, particularly the information stored within it. Security techniques must continuously adapt to dynamically changing conditions in the environment, such as new protocols, applications, and user behaviors. We present a hybrid modular system that utilizes SARIMA and ANFIS algorithms, as well as One Class classifier structures based on convolutional networks, designed to detect anomalies using network user profiles built from frewall event data. This model has been adapted for recognizing and classifying network data for 280 different users, achieving a high accuracy of 78%. Unlike traditional solutions, the presented system combines the analysis of both linear and nonlinear patterns. The information sources include events containing data on visited URLs, website categories, activity time, protocols, and the amount of transmitted data. The integration of the SARIMA model, responsible for capturing data trends, with the adaptive ANFIS system enables more accurate forecasting and identification of user behaviors. The system operates in real-time, allowing for rapid response to potential threats and enhancing network security by dynamically adapting to changes in user activity. This solution can be applied in various environments where continuous monitoring and user authorization are crucial; from corporate internal networks to public organizations with specific data security requirements. Our research demonstrates that the ARIMA-ANFIS hybrid model is an effective tool for behavioral profiling, integrating large-scale data analysis and the scalability needed to manage dynamically evolving network environments.

We present an innovative method for constructing ensembles of modular networks designed for data classification, based on novel one-class classifier-type structures equipped with a so-called glial driver. This concept is inspired by recent neurobiological discoveries highlighting the significant impact of glial cells on cognitive processes in the human brain. The proposed solution was implemented and validated under real-world conditions within a Polish government ministry to automate document routing. The process of building one-class classifier models is also novel, as it involves training a single structure with the entire training sequence before converting the trained structure into a one-class classifier model using glial cells. This approach demonstrates a substantial reduction in the training time of the modular system and a significant improvement in its performance.

Analysis of system logs plays an important role in improving the security of IT systems. System logs are records of events and activities that take place in the system, such as logging in, running applications, accessing files, network connections, etc. Analysis of system logs can provide valuable information about user activities and behaviors, detect suspicious or unauthorised activities, and identify potential threats to system security. Analysis of system logs allows system administrators to identify incorrect or unauthorised activities that may indicate attacks or security breaches. By monitoring and analyzing logs, we can quickly detect unwanted activities, such as hacking attempts, suspicious logins, or DDoS attacks. In this chapter, the possibilities of using convolutional neural networks in the tasks of searching for approximate sequences of events in the logs of computer systems allowing for the detection of anomalies will be presented.

We propose a fast retrieval of full-disk images of the Sun magnetograms collected by the Solar Dynamics Observatory spacecraft. The images are produced at a high rate which entails the need to have a method for efficient searching in this vast collection. The method uses a hand-crafted integer-valued short descriptor of magnetograms and fuzzy rules created by boosting meta-learning. Fuzzy rule base is used as a database index. The speed-up comes from the short descriptor and fuzzy rule indexing. The efficiency and accuracy of the method is proved by the experiments.

We present a new approach to anonymizing personal data in text files. In the conducted research, an approach was applied that enables the analysis of sentence sentences with the use of neural networks. Contrary to other currently proposed methods, the presented work analyzes the context of a fragment of the text, which enables the detection of sensitive information not only on the basis of specific words but on the basis of "understanding" the context, such as "mayor of Paris", "son of the CEO" of a specific company. We present a proprietary solution using convolutional networks connected with glial cells, enabling the selection of the optimal size of the CNN network structure.

The article presents a method of document anonymization using generative adversarial neural networks. Unlike other anonymization methods, in the presented work, the anonymization concerns sensitive data in the form of images placed in text documents. Specifically, it is based on the CycleGAN idea and uses the U-Net model as a generator. To train the model we built a dataset with text documents with embedded real-life images, and medical images. The method is characterized by a very high efficiency, which enables the detection of 99.8% of areas where the sensitive image is located.

One of the biggest problems faced by companies is the sudden departure of employees from the company. Such events may even result in a serious paralysis of the functioning of enterprises in the event of resignation from work by people holding significant positions. Therefore, an extremely important issue is to develop techniques that will allow detecting the planned resignation of a given employee well in advance. Gaining knowledge about the factors influencing this type of events may allow for taking actions aimed at counteracting them. This work proposes a proprietary method based on the use of artificial neural networks to predict employees leaving work and to indicate which of the possible analyzed reasons are the most significant. Ultimately, the proposed system achieved an efficiency of 74 %.

The paper proposes a method for classifying and fast retrieving full-disk images of the Sun chromosphere and corona collected by the Solar Dynamics Observatory spacecraft. The method uses a convolutional autoencoder to encode the solar images in the form of a concise semantic hash. The speed of the retrieval comes from the boosting meta-learning to construct a set of fuzzy rules describing the hash parameters. The rules constitute a set of weak classifiers voting for the final image class. This allows for fast retrieving similar images in vast collections of solar images.

The paper proposes a method for classifying and fast retrieving images which uses boosting metalearning to search for the most salient image features. We use local image keypoints as image features. We construct by boosting a set fuzzy rules describing image feature parameters. The rules constitute a set of weak classifiers voting for the final image class. The method can use various image features, engineered and learned by deep learning methods. We checked the methods on some real-world images.

Along with the development of the Internet, methods of fraud and ways to obtain important data such as logins and passwords or personal sensitive data have evolved. One way of obtaining such information is to impersonate a page the user knows. Such a site usually does not provide any services other than collecting sensitive information from the user. In this paper, we present a way to detect such malicious URL addresses with almost 100% accuracy using convolutional neural networks. Contrary to the previous works, where URL or traffic statistics or web content are analysed, we analyse only the URL text. Thus, the method is faster and detects zero-day attacks. The network we present is appropriately optimised so that it can be used even on mobile devices without significantly affecting its performance.

The significant development of classifiers has made object detection and classification by using neural networks more effective and more straightforward. Unfortunately, there are images where these operations are still difficult due to the overlap of objects or very blurred contours. An example is images obtained from various microscopes, where bacteria or other biological structures can merge, or even have different shapes. To this end, we propose a novel solution based on convolutional auto-encoders and additional two-dimensional image processing techniques to achieve better efficiency in the detection and classification of small objects in such images. In our research, we have included elements such as very weak contours of shapes that may result from the merging of biological objects. The presented method was compared with others, such as a faster recurrent convolutional neural network to indicate the advantages of the proposed solution.

Security of computer systems is now a critical and evolving issue. Current trends try to use behavioural biometrics for continuous authorization. Our work is intended to strengthen network user authentication by a software interaction analysis. In our research, we use HTTP request (URLs) logs that network administrators collect. We use a set of full-convolutional autoencoders and one authentication (one-class) convolutional neural network. The proposed method copes with extensive data from many users and allows to add new users in the future. Moreover, the system works in a real-time manner, and the proposed deep learning framework can use other user behaviour- and software interaction-related features.

We present the possibilities of using convolutional and convolutional recurrent network structures to classify large text sets on the example of job offer descriptions. In the case of recruitment agencies and job offer web pages, it is essential to have a consistent database of offers. Unfortunately, different employers use very different names for the same positions and various descriptions. In this article, we classify job offer texts using feedforward and recurrent convolutional neural networks. We present also two ways to input text data for neural networks. The research was based on five randomly selected work positions. Based on the analyzed texts, the considered neural networks can recognize the type of position with relatively high accuracy.

Sequential pattern mining can discover many interesting phenomena such as bank transactions, web page requesting sequences, customer behavior, etc. There have been many frequent itemset mining algorithms proposed so far, yet it is still a challenging task. In this paper, we propose a deep learning architecture for discovering closed sequences. The U-Net network is trained with random, synthetic sequences and, afterward, is able to discover unknown (not seen during training) sequences. The proposed solution is faster than traditional sequential data mining methods for longer sequences.

In this paper we propose a novel method for invariant image reconstruction with the properly selected degree of symmetry. We make use of Zernike radial moments to represent an image due to their invariance properties to isometry transformations and the ability to uniquely represent the salient features of the image. The regularized ridge regression estimation strategy under symmetry constraints for estimating Zernike moments is proposed. This extended regularization problem allows us to enforces the bilateral symmetry in the reconstructed object. This is achieved by the proper choice of two regularization parameters controlling the level of reconstruction accuracy and the acceptable degree of symmetry. As a byproduct of our studies we propose an algorithm for estimating an angle of the symmetry axis which in turn is used to determine the possible asymmetry present in the image. The proposed image recovery under the symmetry constraints model is tested in a number of experiments involving image reconstruction and symmetry estimation.

Scheduling can be described as a decision-making process. It is applied in various applications, such as manufacturing, airports, and information processing systems. More so, the presence of symmetry is common in certain types of scheduling problems. There are three types of parallel machine scheduling problems (PMSP): uniform, identical, and unrelated parallel machine scheduling problems (UPMSPs). Recently, UPMSPs with setup time had attracted more attention due to its applications in different industries and services. In this study, we present an efficient method to address the UPMSPs while using a modified harris hawks optimizer (HHO). The new method, called MHHO, uses the salp swarm algorithm (SSA) as a local search for HHO in order to enhance its performance and to decrease its computation time. To test the performance of MHHO, several experiments are implemented using small and large problem instances. Moreover, the proposed method is compared to several state-of-art approaches used for UPMSPs. The MHHO shows better performance in both small and large problem cases.

This work contains a solution for game metrics analysis based on a visual data stream dedicated for the player. The solution does not require interference in the programming code of the analyzed game and it is only based on image processing. It is possible to analyze several aspects of the game simultaneously, for example health/energy bars, current weapon used, number of objects worn (aid kits, ammunition). There have been presented methods using cascading classifiers and their training to detect the desired objects on the screen and to prepare data for other stages of processing, e.g. OCR. The effect of the methods is a gameplay chart that allows a thorough analysis of the player's actions in the game world and his or her advancement. The solution is fast enough that it can be used not only in previously recorded gameplay analysis, but also in real time during simultaneous gameplay.