Network Learning Research Articles

MBI-TCMR: A TRADITIONAL CHINESE FORMULA RECOMMENDATION DEEP LEARNING MODEL BASED ON MULTI-TASK META LEARNING AND BIOLOGICAL NETWORK STRUCTURE

The performance of existing traditional Chinese medicine (TCM) recommendation models is generally poor because of their weak generalization ability, overfitting, and inability to use known biological networks. Therefore, building a TCM recommendation model based on artificial intelligence has currently become an important bioinformatics task. This study aimed to design a multitask meta-learning model with good biological interpretation for TCM formula recommendation (MBI-TCMR) for deep learning regularization. This method was based on the known biological network structure to sparse the deep learning network, solve the overfitting problem of the model, and enhance the biological interpretability of the model. Furthermore, a multi-learning framework based on meta-learning was also proposed. The framework allowed the MBI-TCMR model to mine knowledge of TCM formulas and quickly adapt to different types of TCM formula recommendation tasks. Finally, we used a gradient-based deep learning feature backtracking method to calculate the feature weight for each neuron. This weight could provide valuable explanatory information for researchers to study how the model made its medicine recommendations. We designed three independent experiments. The experimental results showed that the hit ratio (HR), AUC, and recall and precision value of the MBI-TCMR model outperformed the existing TCM formula recommendation models. The MBI-TCMR model’s HR of top 1–10 reached 0.15–0.9 (Gynecologic Disease Dataset). HR was 10 for the MBI-TCMR model, which was an improvement of 11.1% compared with the best baseline model. The bio-enrichment analysis showed that the model exhibited good bio-interpretation. In summary, this study proposed a novel TCM formula recommendation model, which expanded the application of the artificial intelligence model and achieved good results.

Cross-shaped windows transformer with self-supervised pretraining for clinically significant prostate cancer detection in bi-parametric MRI.

Bi-parametric magnetic resonance imaging (bpMRI) has demonstrated promising results in prostate cancer (PCa) detection. Vision transformers have achieved competitive performance compared to convolutional neural network (CNN) in deep learning, but they need abundant annotated data for training. Self-supervised learning can effectively leverage unlabeled data to extract useful semantic representations without annotation and its associated costs. This study proposes a novel self-supervised learning framework and a transformer model to enhance PCa detection using prostate bpMRI. We introduce a novel end-to-end Cross-Shaped windows (CSwin) transformer UNet model, CSwin UNet, to detect clinically significant prostate cancer (csPCa) in prostate bpMRI. We also propose a multitask self-supervised learning framework to leverage unlabeled data and improve network generalizability. Using a large prostate bpMRI dataset (PI-CAI) with 1476 patients, we first pretrain CSwin transformer using multitask self-supervised learning to improve data-efficiency and network generalizability. We then finetune using lesion annotations to perform csPCa detection. We also test the network generalization using a separate bpMRI dataset with 158 patients (Prostate158). Five-fold cross validation shows that self-supervised CSwin UNet achieves 0.888±0.010 aread under receiver operating characterstics curve (AUC) and 0.545±0.060 Average Precision (AP) on PI-CAI dataset, significantly outperforming four comparable models (nnFormer, Swin UNETR, DynUNet, Attention UNet, UNet). On model generalizability, self-supervised CSwin UNet achieves 0.79 AUC and 0.45 AP, still outperforming all other comparable methods and demonstrating good generalization to external data. This study proposes CSwin UNet, a new transformer-based model for end-to-end detection of csPCa, enhanced by self-supervised pretraining to enhance network generalizability. We employ an automatic weighted loss (AWL) to unify pretext tasks, improving representation learning. Evaluated on two multi-institutional public datasets, our method surpasses existing methods in detection metrics and demonstrates good generalization to external data.

Analysis of athletes’ technical action based on deep learning

In order to raise athletes’ technical proficiency and overall performance, this paper uses deep learning technology to precisely assess table tennis technical activities. The paper builds a hybrid neural network model for technical action analysis of table tennis players, based on Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) in deep learning. First, CNN extracts the spatial characteristics from the video frames. After that, an LSTM processes these data in a time series to accurately recognize and classify the movements of athletes. Lastly, through trials, the model is trained and evaluated using the Table Tennis Tracking Network (TTNET) dataset. The experimental findings demonstrate that: (1) In table tennis technical action analysis, the suggested deep learning model performs better than other comparative models. The efficacy of combining convolution with sequence learning is fully demonstrated by the CNN-LSTM hybrid model, which performs best in all indicators when compared to the CNN, LSTM, Multi-Objective Function (MOF), and Convolutional Neural Network—Long Short-Term Memory (CNN-LSTM) combined models. Its accuracy rate, precision rate, recall rate, and F1 score are 0.923, 0.918, 0.925, and 0.921, respectively. In contrast, the performances of LSTM and CNN are also excellent, but the performance of MOF model is relatively low; (2) In the classification of technical actions, the model has the highest classification accuracy of service, reaching 0.930, and the classification accuracy of other technical actions such as forehand stroke, backhand stroke and spike is also above 0.9, and the time sequence consistency index also maintains a high level, indicating that the model can effectively identify and analyze table tennis technical actions. In addition, the performance evaluation of real-time feedback shows that the model can achieve low processing time and feedback delay in video data processing with different lengths, which ensures the real-time and reliability in practical applications. These results show that the proposed model can not only provide accurate technical action recognition, but also provide timely and effective feedback in practical application, which has high practical value. The results of this paper prove the potential of deep learning technology in the analysis of athletes’ technical actions, and provide scientific basis and effective tools for the technical training and optimization of table tennis players.

Multi-modal transcriptomics: integrating machine learning and convolutional neural networks to identify immune biomarkers in atherosclerosis

BackgroundAtherosclerosis, a complex chronic vascular disorder with multifactorial etiology, stands as the primary culprit behind consequential cardiovascular events, imposing a substantial societal and economic burden. Nevertheless, our current understanding of its pathogenesis remains imprecise. In this investigation, our objective is to establish computational models elucidating molecular-level markers associated with atherosclerosis. This endeavor involves the integration of advanced machine learning techniques and comprehensive bioinformatics analyses.Materials and methodsOur analysis incorporated data from three publicly available the Gene Expression Omnibus (GEO) datasets: GSE100927 (104 samples, 30,558 genes), which includes atherosclerotic lesions and control arteries from carotid, femoral, and infra-popliteal arteries of deceased organ donors; GSE43292 (64 samples, 23,307 genes), consisting of paired carotid endarterectomy samples from 32 hypertensive patients, comparing atheroma plaques and intact tissues; and GSE159677 (30,498 single cells, 33,538 genes), examining single-cell transcriptomes of calcified atherosclerotic core plaques and adjacent carotid artery tissues from patients undergoing carotid endarterectomy. Utilizing single-cell sequencing, highly variable atherosclerotic monocyte subpopulations were systematically identified. We analyzed cellular communication patterns with temporal dynamics. The bioinformatics approach Weighted Gene Co—expression Network Analysis (WGCNA) identified key modules, constructing a Protein-Protein Interaction (PPI) network from module-associated genes. Three machine-learning models derived marker genes, formulated through logistic regression and validated via convolutional neural network(CNN) modeling. Subtypes were clustered based on Gene Set Variation Analysis (GSVA) scores, validated through immunoassays.ResultsThree pivotal atherosclerosis-associated genes—CD36, S100A10, CSNK1A1—were unveiled, offering valuable clinical insights. Profiling based on these genes delineated two distinct isoforms: C2 demonstrated potent microbicidal activity, while C1 engaged in inflammation regulation, tissue repair, and immune homeostasis. Molecular docking analyses explored therapeutic potential for Estradiol, Zidovudine, Indinavir, and Dronabinol for clinical applications.ConclusionThis study introduces three signature genes for atherosclerosis, shaping a novel paradigm for investigating clinical immunological medications. It distinguishes the high biocidal C2 subtype from the inflammation-modulating C1 subtype, utilizing identified signature gene as crucial targets.

Advancing Healthcare: Intelligent Speech Technology for Transcription, Disease Diagnosis, and Interactive Control of Medical Equipment in Smart Hospitals

Intelligent Speech Technology (IST) is revolutionizing healthcare by enhancing transcription accuracy, disease diagnosis, and medical equipment control in smart hospital environments. This study introduces an innovative approach employing federated learning with Multi-Layer Perceptron (MLP) and Gated Recurrent Unit (GRU) neural networks to improve IST performance. Leveraging the “Medical Speech, Transcription, and Intent” dataset from Kaggle, comprising a variety of speech recordings and corresponding medical symptom labels, noise reduction was applied using a Wiener filter to improve audio quality. Feature extraction through MLP and sequence classification with GRU highlighted the model’s robustness and capacity for detailed medical understanding. The federated learning framework enabled collaborative model training across multiple hospital sites, preserving patient privacy by avoiding raw data exchange. This distributed approach allowed the model to learn from diverse, real-world data while ensuring compliance with strict data protection standards. Through rigorous five-fold cross-validation, the proposed Fed MLP-GRU model demonstrated an accuracy of 98.6%, with consistently high sensitivity and specificity, highlighting its reliable generalization across multiple test conditions. In real-time applications, the model effectively performed medical transcription, provided symptom-based diagnostic insights, and facilitated hands-free control of healthcare equipment, reducing contamination risks and enhancing workflow efficiency. These findings indicate that IST, powered by federated neural networks, can significantly improve healthcare delivery, accuracy in patient diagnosis, and operational efficiency in clinical settings. This research underscores the transformative potential of federated learning and advanced neural networks for addressing pressing challenges in modern healthcare and setting the stage for future innovations in intelligent medical technology.

Prioritization of Software Bugs Using Entropy‐Based Measures

ABSTRACTOpen‐source software is evolved through the active participation of users. In general, a user request for bug fixing, the addition of new features, and feature enhancements. Due to this, the software repositories are increasing day by day at an enormous rate. Additionally, user distinct requests add uncertainty and irregularity to the reported bug data. The performance of machine learning algorithms drastically gets influenced by the inappropriate handling of uncertainty and irregularity in the bug data. Researchers have used machine learning techniques for assigning priority to the bug without considering the uncertainty and irregularity in reported bug data. In order to capture the uncertainty and irregularity in the reported bug data, the summary entropy–based measure in combination with the severity and summary weight is considered in this study to predict the priority of bugs in the open‐source projects. Accordingly, the classifiers are build using these measures for different machine learning techniques, namely, k‐nearest neighbor (KNN), naïve Bayes (NB), J48, random forest (RF), condensed nearest neighbor (CNN), multinomial logistic regression (MLR), decision tree (DT), deep learning (DL), and neural network (NNet) for bug priority prediction This research aims to systematically analyze the summary entropy–based machine learning classifiers from three aspects: type of machine learning technique considered, estimation of various performance measures: Accuracy, Precision, Recall, and F‐measure and through existing model comparison. The experimental analysis is carried out using three open‐source projects, namely, Eclipse, Mozilla, and OpenOffice. Out of 145 cases (29 products X 5 priority levels), the J48, RF, DT, CNN, NNet, DL, MLR, and KNN techniques give the maximum F‐measure for 46, 35, 28, 11, 15, 4, 3, and 1 cases, respectively. The result shows that the proposed summary entropy–based approach using different machine learning techniques performs better than without entropy‐based approach and also entropy‐based approach improves the Accuracy and F‐measure as compared with the existing approaches. It can be concluded that the classifier build using summary entropy measure significantly improves the machine learning algorithms' performance with appropriate handling of uncertainty and irregularity. Moreover, the proposed summary entropy–based classifiers outperform the existing models available in the literature for predicting bug priority.

MCGAN-a cutting edge approach to real time investigate of multimedia deepfake multi collaboration of deep generative adversarial networks with transfer learning.

The proliferation of multimedia-based deepfake content in recent years has posed significant challenges to information security and authenticity, necessitating the use of methods beyond dependable dynamic detection. In this paper, we utilize the powerful combination of Deep Generative Adversarial Networks (GANs) and Transfer Learning (TL) to introduce a new technique for identifying deepfakes in multimedia systems. Each of the GAN architectures may be customized to detect subtle changes in different multimedia formats by combining their advantages. A multi-collaborative framework called "MCGAN" is developed because it contains audio, video, and image files. This framework is compared to other state-of-the-art techniques to estimate the overall fluctuation based on performance, improving the accuracy rate by up to 17.333% and strengthening the deepfake detection hierarchy. In order to accelerate the training process overall and enable the system to respond rapidly to novel patterns that indicate deepfakes, TL employs the pre-train technique on the same databases. When it comes to identifying the contents of deepfakes, the proposed method performs quite well. In a range of multimedia scenarios, this enhances real-time detection capabilities while preserving a high level of accuracy. A progressive hierarchy that ensures information integrity in the digital world and related research is taken into consideration in this development.

Strip and boundary detection multi-task learning network for segmentation of meibomian glands.

Automatic segmentation of meibomian glands in near-infrared meibography images is basis of morphological parameter analysis, which plays a crucial role in facilitating the diagnosis of meibomian gland dysfunction (MGD). The special strip shape and the adhesion between glands make the automatic segmentation of meibomian glands very challenging. A strip and boundary detection multi-task learning network (SBD-MTLNet) based on encoder-decoder structure is proposed to realize the automatic segmentation of meibomian glands. A strip mixed attention module (SMAM) is proposed to enhance the network's ability to recognize the strip shape of glands. To alleviate the problem of adhesion between glands, a boundary detection auxiliary network (BDA-Net) is proposed, which introduces boundary features to assist gland segmentation. A self-adaptive interactive information fusion module (SIIFM) based on reverse attention mechanism is proposed to realize information complementation between meibomian gland segmentation and boundary detection tasks. The proposed SBD-MTLNet has been evaluated on an in-house dataset (453 images) and a public dataset MGD-1K (1000 images). Due to the limited number of images, a five-fold cross validation strategy is adopted. Average dice coefficient of the proposed SBD-MTLNet reaches 81.08% and 84.32% on the in-house dataset and the public one, respectively. Comprehensive experimental results demonstrate the effectiveness the proposed SBD-MTLNet, outperforming other state-of-the-art methods. The proposed SBD-MTLNet can focus more on the shape characteristics of the meibomian glands and the boundary contour information between the adjacent glands via multi-task learning strategy. The segmentation results of the proposed method can be used for the quantitative morphological characteristics analysis of meibomian glands, which has potential for the auxiliary diagnosis of MGD in clinic.

Research on Non-Intrusive Load Disaggregation Technology Based on VMD–Nyströmformer–BiTCN

Non-intrusive load disaggregation is a technique that monitors the total electrical load of an entire building or household. It uses a single power metering device to measure the total load. Then, it employs algorithms to break it down into the individual usage of different electrical devices. To address issues in load disaggregation models such as long training times, feature interference caused by the activation of other loads, and accuracy deficiencies caused by behavioral interference from users’ electricity usage habits, this paper proposes a VMD–Nyströmformer–BiTCN network architecture. The variational mode decomposition (VMD) filters the raw power data, reducing errors caused by noise and enhancing the accuracy of decomposing the load. A deep learning network utilizes a modified attention model, Nyströmformer, to reduce feature entanglement and accuracy degradation caused by habitual behavior interference during load disaggregation, while ensuring precise accuracy and improving network operational speed. The training network uses a bidirectional temporal convolutional network (BiTCN) and incorporates a residual network to expand the receptive field, allowing it to receive longer load sequence data and acquire more effective load information, thereby improving the disaggregation effectiveness for target appliances.

Cell Zooming With Energy‐Efficient Offloading in Cellular Network: An Optimization‐Assisted Model

ABSTRACTWith the speedy development of cellular users, conventional mobile network communications could not satisfy the requirements of rising traffic. The design of the user association approach should have the ability to balance the energy saving and traffic hotspot providence from the traffic cell sleeping system. Nevertheless, with the gradually rising denser deployment of small cells (SCs), the HetNets face novel challenges. This work aims to introduce a cell zooming (CZ) scheme that admits the EE traffic offloading. Thereby, the given model includes two phases, (i) energy efficient offloading: In this phase, appropriate traffic flow is predicted for the hybrid deep learning model that includes deep Q network (DQN) and reinforcement learning (RL). This prediction is related to the offload traffic from macrocell to small cell based on constraints like QoS and Energy, respectively. (ii) Cell zooming: After the prediction, adaptive CZ is preceded by an improved CZ factor with optimal threshold selection. For optimization, a new scaled bald eagle search optimization (SBESO) is used. The betterment of SBESO is established on throughput, SINR, and so forth.

Wi-Fi Sensing Techniques for Human Activity Recognition: Brief Survey, Potential Challenges, and Research Directions

Recent advancements in wireless communication technologies have made Wi-Fi signals indispensable in both personal and professional settings. The utilization of these signals for Human Activity Recognition (HAR) has emerged as a cutting-edge technology. By leveraging the fluctuations in Wi-Fi signals for HAR, this approach offers enhanced privacy compared to traditional visual surveillance methods. The essence of this technique lies in detecting subtle changes when Wi-Fi signals interact with the human body, which are then captured and interpreted by advanced algorithms. This paper initially provides an overview of the key methodologies in HAR and the evolution of non-contact sensing, introducing sensor-based recognition, computer vision, and Wi-Fi signal-based approaches, respectively. It then explores tools for Wi-Fi-based HAR signal collection and lists several high-quality datasets. Subsequently, the paper reviews various sensing tasks enabled by Wi-Fi signal recognition, highlighting the application of deep learning networks in Wi-Fi signal detection. The fourth section presents experimental results that assess the capabilities of different networks. The findings indicate significant variability in the generalization capacities of neural networks and notable differences in test accuracy for various motion analyses.

Real-time Mask Detection with YOLOv3 and Ensemble Residual Networks

The proposed system introduced an advanced face mask detections framework leveraging Adaptively YOLOv3 with Hybrid Atrous Convolution-depend Face Detector, optimized through an Ensembles residual Learning Network. This approach aims to enhance the accuracy level and efficacy of face mask detection in real-time situation. By integrating of the adaptive capability of YOLOv3 on the spatial benefits of atrous convolution defined, the system effectively captured varying of facial features and masks placements. The ensemble residual learning further refined detection by amalgamating of multiple learning models, by reducing false positives values and negatives. This hybrid model is meticulous optimized to maintaining of higher detection precision rate and recall, for ensuring robust performance even over diverse and complex conditions. The system’s efficacy will underscored by reduces the inference time and enhanced processing throughput rate, by making it highly suitable for large-scale deployment on public safety applications. By rigorously testing phase, the proposed system demonstrated significant improvements in detection accuracy rate, inference speed, and overall reliability, positioning it will be a pivotal tool in enforcing healthy safety protocols on various public and private settings.

Unsupervised Ground Roll Attenuation via Implicit Neural Representations

Coherent noise attenuation presents a significant challenge compared to incoherent noise, particularly concerning ground roll in land seismic data. Traditional methods encounter limitations with the substantial overlap of ground roll and reflections, resulting in a compromise between signal distortion and residual noise. Recent deep learning strategies, while promising, predominantly rely on supervised learning, necessitating extensive paired training#xD;data, which is frequently scarce in real-world scenarios. Unsupervised approaches often struggle with convergence issues or excessive parameter tuning. We introduce an unsupervised deep learning framework for separating reflections from ground roll to address these hurdles. Leveraging the inherent low-frequency bias of implicit neural representations (INR), our method prioritizes extracting self-similarity features during training. For seismic data, the network learns self-similarity flattened events before those with deep dips and other incoherent noise. To enhance the self-similarity of reflections, we first apply normal moveout (NMO) correction to flatten reflections and then use the network to extract these flattened reflections from the NMO-corrected data. Moreover, to ensure convergence, we integrate a horizontal derivative regularization term into the loss function, penalizing horizontal variations. This regularization prevents the extraction of unflattened events through extended training, ensuring convergence with high fidelity of flattened reflections. It additionally streamlines parameter tuning, yielding stable outputs and obviating the need for early stopping. We validate our method using synthetic and real land data examples, comparing it with traditional f-k filters with real data and demonstrating its superiority in noise attenuation and signal preservation.

Artificial intelligence improves the identification of fur seals recorded at Southern Brazilian coast

The genus Arctocephalus represents the group of fur seals that mainly inhabit the Southern Hemisphere. In general, Arctocephalus species are extremely similar in appearance, often making it very difficult to impossible to distinguish them only by characteristics of their external morphology. In this context, it is important to find new tools to differentiate them, especially in locations outside of their traditional distribution area, such as Brazilian waters, in order to take appropriate actions for their management. This study proposes the use of an artificial intelligence method, based on machine learning and convolutional neural networks, to classify and identify three species of southern fur seals by analysing 121 facial images from living specimens of Arctocephalus australis, A. gazella, and A. tropicalis found on the Brazilian coast. The image database Keywords: citizen science convolutional neural networks, machine learning, pinnipeds.

A multi-relational neighbors constructed graph neural network for heterophily graph learning

A multi-relational neighbors constructed graph neural network for heterophily graph learning

Etiology of Late-Onset Alzheimer’s Disease, Biomarker Efficacy, and the Role of Machine Learning in Stage Diagnosis

Late-onset Alzheimer’s disease (LOAD) is a subtype of dementia that manifests after the age of 65. It is characterized by progressive impairments in cognitive functions, behavioral changes, and learning difficulties. Given the progressive nature of the disease, early diagnosis is crucial. Early-onset Alzheimer’s disease (EOAD) is solely attributable to genetic factors, whereas LOAD has multiple contributing factors. A complex pathway mechanism involving multiple factors contributes to LOAD progression. Employing a systems biology approach, our analysis encompassed the genetic, epigenetic, metabolic, and environmental factors that modulate the molecular networks and pathways. These factors affect the brain’s structural integrity, functional capacity, and connectivity, ultimately leading to the manifestation of the disease. This study has aggregated diverse biomarkers associated with factors capable of altering the molecular networks and pathways that influence brain structure, functionality, and connectivity. These biomarkers serve as potential early indicators for AD diagnosis and are designated as early biomarkers. The other biomarker datasets associated with the brain structure, functionality, connectivity, and related parameters of an individual are broadly categorized as clinical-stage biomarkers. This study has compiled research papers on Alzheimer’s disease (AD) diagnosis utilizing machine learning (ML) methodologies from both categories of biomarker data, including the applications of ML techniques for AD diagnosis. The broad objectives of our study are research gap identification, assessment of biomarker efficacy, and the most effective or prevalent ML technology used in AD diagnosis. This paper examines the predominant use of deep learning (DL) and convolutional neural networks (CNNs) in Alzheimer’s disease (AD) diagnosis utilizing various types of biomarker data. Furthermore, this study has addressed the potential scope of using generative AI and the Synthetic Minority Oversampling Technique (SMOTE) for data augmentation.

Neural Networks for Classification of Eye Conjunctivitis in Telehealth: a Conceptual Architecture

Telehealth systems have developed rapidly into more conventional ways that can provide medical assistance, especially for people in remote areas. Despite rapid technological and practical developments, there are still many knowledge gaps regarding the effective use of telemedicine. Annually, nearly 1-4% of the general population might experience conjunctivitis. This study is focused on an experimental design for the classification of degrees of severity in colour medical images in telemedicine, in particular red as one of the key symptoms in the diagnosis of various pathologies. The quality of digital images is a pivotal thing in terms of telemedicine for accurate diagnosis because degraded or distorted colours can lead to errors. This study focused on the use of digital images in teleconsultation, in particular images displaying conjunctivitis (red eyes) as a case study since this pathology integrates red in its diagnosis. The deep self-organising map is suggested to be applied to classify the different severities. Moreover, U-Net, a deep learning network, is proposed to employ the segmentation of eye images for better feature extraction. Although this approach is focused on the problem of red eye image classification, it can be extended in the future to also be applied to other pathologies.

EfficientNet-resDDSC: A Hybrid Deep Learning Model Integrating Residual Blocks and Dilated Convolutions for Inferring Gene Causality in Single-Cell Data.

Gene Regulatory Networks (GRNs) reveal complex interactions between genes in organisms, crucial for understanding the life system's operation. The rapid development of biotechnology, especially single-cell RNA sequencing (scRNA-seq), has generated a large amount of scRNA-seq data, which can be analyzed to explore the regulatory relationships between genes at the single-cell level. Previous models used to construct GRNs mainly aim at constructing associative relationships between genes, but usually fail to accurately reveal the causality between genes. Therefore, we present a hybrid deep learning model called EfficientNet-resDDSC (the EfficientNet with Residual Blocks and Depthwise Separable Dilated Convolutions) to infer causality between genes. The model inherits the basic structure of EfficientNet-B0 and incorporates residual blocks as well as dilated convolutions. The model's ability to extract low-level features at the primary stage is enhanced by introducing residual blocks. The model combines Depthwise Separable Convolution (DSC) in the inverted linear bottleneck layers with the dilated convolutions to expand the model's receptive fields without increasing the computational effort. This design enables the model to comprehensively reveal potential relationships among different genes in high-dimensional and high-noise single-cell data. In comparison with the five existing deep learning network models, EfficientNet-resDDSC's overall performance is significantly better than others on four datasets. In this study, EfficientNet-resDDSC was further applied to construct GRNs for breast cancer patients, focusing on the related regulatory genes of the key gene BRCA1, which contributes to the advancement of breast cancer research and treatment strategies.

«The Trouble with Bias» in Neural Networks: Conflict and Ethical Challenges

This article is devoted to the analysis of ethical and conflict challenges related to the trouble with bias in neural networks. The necessity of a correct, scientifically based explication of the phenomenon of bias is postulated in order to build models correcting this problem as a necessary element of the software development process based on artificial intelligence algorithms. The history of the development of neural networks is considered from the origin of the idea of a mechanical organism to the construction of modern models of an artificial neuron. The most significant characteristics of modern neural networks are highlighted: architecture, weights and offsets, activation functions, inferences, and learning methods.A detailed description of natural language as a neural network learning resource is given, and programming in natural language is analyzed. The specificity of the natural language of the neural network as a set of linguistic practices reflecting the entire digitized experience of mankind, including stereotypes, inequalities, hate speech and other phenomena, ultimately producing the trouble with bias, is emphasized.Considerable attention is paid to the analysis of the phenomena of “politics classification”, “power discourse”, “cultural violence” in the context of the search for methodological foundations of natural language filtering and censorship strategies in the process of constructing a neural network.Separately, it is emphasized how the errors in neural networks are reflected in the existing ethical and conflict studies debates around the problem of artificial intelligence. It is concluded that the current assessment of the moral aspects of the problem does not imply granting neural networks the status of a moral agent and places the ethical expertise of the product on its developers. It is particularly noted that the conflict aspect of the trouble with bias lies in its recognition exclusively in relation to groups that have now acquired the “sensitive” status of discriminated against as a result of social conflicts.In conclusion, the paper substantiates the urgent need to optimize artificial intelligence in order to reduce the destructive potential of the trouble with bias, which necessarily implies the modification of social relations in the broader context of the struggle of excluded groups for the right to be recognized as discriminated against.

Conformalized Graph Learning for Molecular ADMET Property Prediction and Reliable Uncertainty Quantification.

Drug discovery and development is a complex and costly process, with a substantial portion of the expense dedicated to characterizing the absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties of new drug candidates. While the advent of deep learning and molecular graph neural networks (GNNs) has significantly enhanced in silico ADMET prediction capabilities, reliably quantifying prediction uncertainty remains a critical challenge. The performance of GNNs is influenced by both the volume and the quality of the data. Hence, determining the reliability and extent of a prediction is as crucial as achieving accurate predictions, especially for out-of-domain (OoD) compounds. This paper introduces a novel GNN model called conformalized fusion regression (CFR). CFR combined a GNN model with a joint mean-quantile regression loss and an ensemble-based conformal prediction (CP) method. Through rigorous evaluation across various ADMET tasks, we demonstrate that our framework provides accurate predictions, reliable probability calibration, and high-quality prediction intervals, outperforming existing uncertainty quantification methods.