All issues

Interpretability in recent deep learning models has become an epicenter of research particularly in sensitive domains such as healthcare, and finance. Concept bottleneck models have emerged as a promising approach for achieving transparency and interpretability by leveraging a set of humanunderstandable concepts as an intermediate representation before the prediction layer. However, manual concept annotation is discouraged due to the time and effort involved. Our work explores the potential of large language models (LLMs) for generating high-quality concept banks and proposes a multimodal evaluation metric to assess the quality of generated concepts. We investigate three key research questions: the ability of LLMs to generate concept banks comparable to existing knowledge bases like ConceptNet, the sufficiency of unimodal text-based semantic similarity for evaluating concept-class label associations, and the effectiveness of multimodal information in quantifying concept generation quality compared to unimodal concept-label semantic similarity. Our findings reveal that multimodal models outperform unimodal approaches in capturing concept-class label similarity. Furthermore, our generated concepts for the CIFAR-10 and CIFAR-100 datasets surpass those obtained from ConceptNet and the baseline comparison, demonstrating the standalone capability of LLMs in generating highquality concepts. Being able to automatically generate and evaluate high-quality concepts will enable researchers to quickly adapt and iterate to a newer dataset with little to no effort before they can feed that into concept bottleneck models.

Correct modeling of complex dynamics of urban transportation flows requires the collection of large volumes of empirical data to specify types of the modes and their identification. At the same time, setting a large number of observation posts is expensive and technically not always feasible. All this results in insufficient factographic support for the traffic control systems as well as for urban planners with the obvious consequences for the quality of their decisions. As one of the means to provide large-scale data collection at least for the qualitative situation analysis, the wide-area video cameras are used in different situation centers. There they are analyzed by human operators who are responsible for observation and control. Some video cameras provided their videos for common access, which makes them a valuable resource for transportation studies. However, there are significant problems with getting qualitative data from such cameras, which relate to the theory and practice of image processing. This study is devoted to the practical application of certain mainstream neuro-networking technologies for the estimation of essential characteristics of actual transportation flows. The problems arising in processing these data are analyzed, and their solutions are suggested. The convolution neural networks are used for tracking, and the methods for obtaining basic parameters of transportation flows from these observations are studied. The simplified neural networks are used for the preparation of training sets for the deep learning neural network YOLOv4 which is later used for the estimation of speed and density of automobile flows.

Deep learning’s power stems from complex architectures; however, these can lead to overfitting, where models memorize training data and fail to generalize to unseen examples. This paper proposes a novel probabilistic approach to mitigate this issue. We introduce two key elements: Truncated Log-Uniform Prior and Truncated Log-Normal Variational Approximation, and Automatic Relevance Determination (ARD) with Bayesian Deep Neural Networks (BDNNs). Within the probabilistic framework, we employ a specially designed truncated log-uniform prior for noise. This prior acts as a regularizer, guiding the learning process towards simpler solutions and reducing overfitting. Additionally, a truncated log-normal variational approximation is used for efficient handling of the complex probability distributions inherent in deep learning models. ARD automatically identifies and removes irrelevant features or weights within a model. By integrating ARD with BDNNs, where weights have a probability distribution, we achieve a variational bound similar to the popular variational dropout technique. Dropout randomly drops neurons during training, encouraging the model not to rely heavily on any single feature. Our approach with ARD achieves similar benefits without the randomness of dropout, potentially leading to more stable training.

To evaluate our approach, we have tested the model on two datasets: the Canadian Institute For Advanced Research (CIFAR-10) for image classification and a dataset of Macroscopic Images of Wood, which is compiled from multiple macroscopic images of wood datasets. Our method is applied to established architectures like Visual Geometry Group (VGG) and Residual Network (ResNet). The results demonstrate significant improvements. The model reduced overfitting while maintaining, or even improving, the accuracy of the network’s predictions on classification tasks. This validates the effectiveness of our approach in enhancing the performance and generalization capabilities of deep learning models.

The reconstruction of charged particle trajectories in tracking detectors is a key problem in the analysis of experimental data for high energy and nuclear physics.

The amount of data in modern experiments is so large that classical tracking methods such as Kalman filter can not process them fast enough. To solve this problem, we have developed two neural network algorithms of track recognition, based on deep learning architectures, for local (track by track) and global (all tracks in an event) tracking in the GEM tracker of the BM@N experiment at JINR (Dubna). The advantage of deep neural networks is the ability to detect hidden nonlinear dependencies in data and the capability of parallel execution of underlying linear algebra operations.

In this work we generalize these algorithms to the cylindrical GEM inner tracker of BESIII experiment. The neural network model RDGraphNet for global track finding, based on the reverse directed graph, has been successfully adapted. After training on Monte Carlo data, testing showed encouraging results: recall of 98% and precision of 86% for track finding.

The local neural network model TrackNETv2 was also adapted to BESIII CGEM successfully. Since the tracker has only three detecting layers, an additional neuro-classifier to filter out false tracks have been introduced. Preliminary tests demonstrated the recall value at the first stage of 99%. After applying the neuro-classifier, the precision was 77% with a slight decrease of the recall to 94%. This result can be improved after the further model optimization.

During the last several decades, cardiovascular disease has surpassed all others as the leading cause of mortality in both high-income and low-income countries. The mortality rate from heart disorders may be lowered with early identification and close clinical monitoring. However, it is not feasible to adequately monitor patients every day, and 24-hour consultation with a doctor is not a feasible option, since it requires more sagacity, time, and knowledge than is currently available.

In this study, we examine the Explainable Artificial Intelligence (XAI) technique, namely, the SHAP interpretability approach, in order to educate the medical professionals about the Explainable AI (XAI) methods that can be helpful in healthcare. The XAI methods enhance the trust and understandability of both practitioners and Health Researchers in AI Models. In this work, we propose a composite Deep Learning model: Bi-LSTM+CNN model to effectively predict heart disease from patient data. After balancing the dataset, the Bi-LSTM+CNN model was used. In contrast to other studies, our proposed hybrid deep learning model produced excellent experimental results, including 99.05% accuracy, 99% precision, 99% recall, and 99% F1-score.

This work analyzes the problem of traffic signs recognition in intelligent transport systems. The basic concepts of computer vision and image recognition tasks are considered. The most effective approach for solving the problem of analyzing and recognizing images now is the neural network method. Among all kinds of neural networks, the convolutional neural network has proven itself best. Activation functions such as Relu and SoftMax are used to solve the classification problem when recognizing traffic signs. This article proposes a technology for recognizing traffic signs. The choice of an approach for solving the problem based on a convolutional neural network due to the ability to effectively solve the problem of identifying essential features and classification. The initial data for the neural network model were prepared and a training sample was formed. The Google Colaboratory cloud service with the external libraries for deep learning TensorFlow and Keras was used as a platform for the intelligent system development. The convolutional part of the network is designed to highlight characteristic features in the image. The first layer includes 512 neurons with the Relu activation function. Then there is the Dropout layer, which is used to reduce the effect of overfitting the network. The output fully connected layer includes four neurons, which corresponds to the problem of recognizing four types of traffic signs. An intelligent traffic sign recognition system has been developed and tested. The used convolutional neural network included four stages of convolution and subsampling. Evaluation of the efficiency of the traffic sign recognition system using the three-block cross-validation method showed that the error of the neural network model is minimal, therefore, in most cases, new images will be recognized correctly. In addition, the model has no errors of the first kind, and the error of the second kind has a low value and only when the input image is very noisy.

Predicting novel drug properties is fundamental to polypharmacology, repositioning, and the study of biologically active substances during the preclinical phase. The use of machine learning, including deep learning methods, for the identification of drug – target interactions has gained increasing popularity in recent years.

The objective of this study was to develop a method for recognizing psychotropic effects and drug mechanisms of action (drug – target interactions) based on an analysis of the bioelectrical activity of the brain using artificial intelligence technologies.

Intracranial electroencephalographic (EEG) signals from rats were recorded (4 channels at a sampling frequency of 500 Hz) after the administration of psychotropic drugs (gabapentin, diazepam, carbamazepine, pregabalin, eslicarbazepine, phenazepam, arecoline, pentylenetetrazole, picrotoxin, pilocarpine, chloral hydrate). The signals were divided into 2-second epochs, then converted into $2000\times 4$ images and input into an autoencoder. The output of the bottleneck layer was subjected to classification and clustering using t-SNE, and then the distances between resulting clusters were calculated. As an alternative, an approach based on feature extraction with dimensionality reduction using principal component analysis and kernel support vector machine (kSVM) classification was used. Models were validated using 5-fold cross-validation.

The classification accuracy obtained for 11 drugs during cross-validation was $0.580 \pm 0.021$, which is significantly higher than the accuracy of the random classifier $(0.091 \pm 0.045, p < 0.0001)$ and the kSVM $(0.441 \pm 0.035, p < 0.05)$. t-SNE maps were generated from the bottleneck parameters of intracranial EEG signals. The relative proximity of the signal clusters in the parametric space was assessed.

The present study introduces an original method for biopotential-mediated prediction of effects and mechanism of action (drug – target interaction). This method employs convolutional neural networks in conjunction with a modified selective parameter reduction algorithm. Post-treatment EEGs were compressed into a unified parameter space. Using a neural network classifier and clustering, we were able to recognize the patterns of neuronal response to the administration of various psychotropic drugs.

In our research, we focus on the problem of classification for silent speech recognition to develop a brain– computer interface (BCI) based on electroencephalographic (EEG) data, which will be capable of assisting people with mental and physical disabilities and expanding human capabilities in everyday life. Our previous research has shown that the silent pronouncing of some words results in almost identical distributions of electroencephalographic signal data. Such a phenomenon has a suppressive impact on the quality of neural network model behavior. This paper proposes a data processing technique that distinguishes between statistically remote and inseparable classes in the dataset. Applying the proposed approach helps us reach the goal of maximizing the semantic load of the dictionary used in BCI.

Furthermore, we propose the existence of a statistical predictive criterion for the accuracy of binary classification of the words in a dictionary. Such a criterion aims to estimate the lower and the upper bounds of classifiers’ behavior only by measuring quantitative statistical properties of the data (in particular, using the Kolmogorov – Smirnov method). We show that higher levels of classification accuracy can be achieved by means of applying the proposed predictive criterion, making it possible to form an optimized dictionary in terms of semantic load for the EEG-based BCIs. Furthermore, using such a dictionary as a training dataset for classification problems grants the statistical remoteness of the classes by taking into account the semantic and phonetic properties of the corresponding words and improves the classification behavior of silent speech recognition models.

We found a predominance of α-rhythm patterns in the left hemisphere in healthy people compared to people with COVID-19 history. Moreover, we observe a significant decrease in the left hemisphere contribution to the speech center area in people who have undergone COVID-19 when performing speech tasks.

Our findings show that the signal in healthy subjects is more spatially localized and synchronized between hemispheres when performing tasks compared to people who recovered from COVID-19. We also observed a decrease in low frequencies in both hemispheres after COVID-19.

EEG-patterns of COVID-19 are detectable in an unusual frequency domain. What is usually considered noise in electroencephalographic (EEG) data carries information that can be used to determine whether or not a person has had COVID-19. These patterns can be interpreted as signs of hemispheric desynchronization, premature brain ageing, and more significant brain strain when performing simple tasks compared to people who did not have COVID-19.

In our work, we have shown the applicability of neural networks in helping to detect the long-term effects of COVID-19 on EEG-data. Furthermore, our data following other studies supported the hypothesis of the severity of the long-term effects of COVID-19 detected on the EEG-data of EEG-based BCI. The presented findings of functional activity of the brain– computer interface make it possible to use machine learning methods on simple, non-invasive brain–computer interfaces to detect post-COVID syndrome and develop progress in neurorehabilitation.