The electrocardiogram (ECG) stands out as one of the most frequently used medical tests, playing a crucial role in the accurate diagnosis and treatment of patients. While ECG devices generate a huge amount of data, only a fraction of it holds valuable medical information. To deal with this problem, many compression algorithms and filters have been developed over the years. However, the rapid development of new machine-learning techniques introduces new challenges. To address this class of problems, we have introduced a FlexPoints algorithm. This innovative algorithm searches for characteristic points on the ECG signal and ignores all other points that lack pertinent medical information. The conducted experiments have demonstrated that our proposed algorithm can significantly reduce the number of data points representing ECG signals without losing valuable medical insights. These sparse but essential characteristic points, referred to as flex points, serve as well-fitted input for modern machine learning models. Such models exhibit enhanced performance when using flex points as input, as opposed to raw data or data compressed by other popular algorithms.

Supervised models often suffer from a multitude of possible combinations of hyperparameters, rigid nonadaptive architectures, underfitting, overfitting, the curse of dimensionality, etc. They consume many resources and slow down the optimization process. As real-world objects are related and similar, we should adapt not only network parameters but also network structure to represent patterns and relationships better. When the network reproduces the most essential and frequent data patterns and relationships and aggregate similarities, it becomes not only efficient but also explainable and trustworthy. This paper presents a new approach to detect and represent similarities of numerical training examples to self-adapt a network structure and its parameters. Such a network will facilitate the classification by identifying hyperspace regions associated with the defined classes in a training dataset. Our approach demonstrates its ability to automatically reduce input data dimension by removing features that produce distortions and do not support the classification process. The presented adaptation algorithm uses only a few optional hyperparameters and produces a sparse associative neural network structure that fits contextually any given dataset by detecting data similarities and constructing hypercuboids in data space. The explanation of these associative adaptive techniques is followed by the comparisons of the classification results against other state-of-the-art models and methods.

Modern methods and networks of supervised learning use a vast amount of computational resources when adapting to large datasets. They are unable to incorporate new training examples into trained models quickly and to represent internal knowledge for quick adaptation to other computational tasks without retraining. The human brain is capable of representing and retrieving vast amounts of information and can create associations between its various pieces based on frequent relationships and patterns. The backpropagation algorithm is not the best and only way to train neural networks, especially since its use is limited to feed-forward architectures. Brain structures can be modeled using graph architectures that are not feed-forward but recursive with many feedback connections. This paper introduces Multi-Associative Graph Networks that enable the representation of associated training data and objects transformed from relational databases. These graphs store the data along with the most useful relationships to facilitate computational intelligence processes. We describe the associative transformation algorithm allowing for the transformation of any relational database into this graph network, reproducing stored relationships and enriching them with newly detected ones. We also introduce the tuning algorithm that learns to associate different priorities with different neurons representing objects to improve the relational dependencies and classification results. Finally, we draw conclusions from the comparisons to other state-of-the-art models.

The purpose of this work is to compare structural associative memories with modern Hopfield networks. The paper discusses various properties of structural associative memories. These memories store multi-object scenes and are used in conjunction with deep neural networks to establish the system's semantic memory. The system of storing scenes in associative memory developed in this paper uses images of objects recognized by a deep neural network within the observed scene. The research derives dependencies on the optimal context size required for scene retrieval from memory. Additionally, it demonstrates how introducing inhibition in graph operations can enhance the memory capacity of structural associative memories. Statistical relationships between memory parameters and stored scenes are established to minimize reproduction errors when retrieving scenes from memory. Furthermore, the study reveals that the frequency profile of objects within scenes significantly affects the estimation of recognition errors. The presented method is compared to modern Hopfield networks with extensive memory capacity through several experiments.

Due to social media’s ubiquitousness, most advertising campaigns take place on such platforms as Facebook, Twitter, or Instagram. As a result, Natural Language Processing has become an essential tool to extract information about the users: their personality traits, brand preferences, distinctive vocabulary, etc. Such data can be further used to create text adverts profiled to engage with users who share a certain set of features. While most of the algorithms capable of processing the text are neural network-driven, associative graphs serve the same purpose, attaining usually similar or better accuracy, but being more explainable than black box-like models based on neural networks. This paper presents an associative graph for natural language processing and fine-grained sentiment analysis. The ability of associative graphs to represent complex relations between phrases can be used to create a model capable of classifying the input data into many categories simultaneously with high accuracy and efficiency. This approach enabled us to acquire a model performing similarly or better than the state-of-the-art solutions while being more explicit and easier to create and explain.

Until now, the simultaneous influence of air pollution assessed by measuring the objective marker of exposition (1-hydroxypirene, 1-OHP) and atopy on the development of allergic airway diseases has not been studied. The aim of this study was to determine the pathomechanism of the allergic response to PM2.5 in atopic and non-atopic patients. We investigated the changes in peripheral blood basophil activity of patients after stimulation with the birch pollen allergen alone, the allergen combined with PM2.5 (BP), PM2.5 alone, a concentration of 1-OHP in urine, and a distance of residence from the main road in 30 persons. Activation by dust alone was positive for all concentrations in 83% of atopic and 75% of non-atopic assays. In the group of people with atopy, the simultaneous activation of BP gave a higher percentage of active basophils compared to the sum of activation with dust and birch pollen alone (B + P) for all concentrations. The difference between BP and B + P was 117.5 (p = 0.02) at a PM concentration of 100 μg. Such a relationship was not observed in the control group. The correlation coefficient between the distance of residence from major roads and urinary 1-OHP was 0.62. A Pearson correlation analysis of quantitative variables was performed, and positive correlation results were obtained in the atopy group between BP and 1-OH-P. Exposure to birch pollen and PM2.5 has a synergistic effect in sensitized individuals. The higher the exposure to pollutants, the higher the synergistic basophil response to the allergen and PM in atopic patients.

Fast discovering of various relationships in data is an important feature of modern data mining, cognitive, knowledge-based, and explainable AI systems, including deep neural networks. The ability to represent a rich set of relationships between stored data and objects is essential for fast inferences, finding associations, representing knowledge, and extracting useful patterns or other pieces of information. This paper introduces self-balancing, aggregating, and sorting ASA-graphs for efficient data representation in various data structures, databases, and data mining systems. These graphs are smaller and use more efficient algorithms for searching, inserting, and removing data than the most commonly used self-balancing trees. ASA-graphs also automatically aggregate and count all duplicates of values and represent them by the same nodes, connecting them in order, and simultaneously providing very fast data access based on a binary search tree approach. The proposed ASA-graph structure combines the advantages of sorted lists, binary search trees, B-trees, and B+trees, eliminating their weaknesses. Our experiments proved that the ASA-graphs outperform many commonly used self-balancing trees.