Model Training Research Articles

A Robust Framework for Detecting Brute-Force Attacks through Deep Learning Techniques

A considerable concern arises with the precise identification of brute-force threats within a networked environment. It emphasizes the need for new methods, as existing ones often lead to many false alarms, as well as delays in real-time threat detection. To tackle these issues, this study proposes a novel intrusion detection framework that utilizes deep learning models for more accurate and efficient detection of brute-force attacks. The framework’s structure includes data collection and preprocessing components performed at the outset of the study using the CSE-CICIDS2018 dataset. The design architecture includes data collection and preprocessing steps. Feature extraction and selection techniques are employed to optimize data for model training. Further, after building the model, various attributes are extracted from the data from feature selection to be used in the training. Then, the construction of multiple architectures of deep learning algorithms, which include Artificial Neural Networks (ANN), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Long Short-Term Memory (LSTM) models. Evaluation results show CNN and LSTM achieved the highest accuracy of 99.995 Parsant and 99.99 Parsant respectively. It showcases its ability to detect complex attack patterns in network traffic. It indicates that the CNN network got the best optimum results with a test time of 9.94 seconds. This establishes CNN as an effective method, achieving high accuracy quickly. In comparison, we have surpassed the accuracy of current methods while addressing their weaknesses. The findings are consistent with the effectiveness of CNN in brute-force attack detection frameworks as a more accurate and faster alternative, increasing the capability of detecting intrusions on a network in real-time.

Prediction of gene expression using histone modification patterns extracted by Particle Swarm Optimization.

Histone modifications play an important role in transcription regulation. Although the general importance of some histone modifications for transcription regulation has been previously established, the relevance of others and their interaction is subject to ongoing research. By training Machine Learning models to predict a gene's expression and explaining their decision making process, we can get hints on how histone modifications affect transcription. In previous studies, trained models were either hardly explainable or the models were trained solely on the abundance of histone modifications. Based on other studies, which used histone modification patterns, rather than their abundance, to identify potential regulatory elements, we hypothesize the histone modification pattern in a gene's promoter to be more predictive for gene expression. We used an optimisation algorithm to extract predictive histone modification profiles. Our algorithm called PatternChrome achieved an average AUC score of 0.9029 over 56 samples for binary classification, outperforming all previous algorithms for the same task. We explained the models decisions to deduce the effect of specific features, certain histone modifications or promoter positions on transcription regulation. Although the predictive histone modification patterns were extracted for each sample separately, they can be used to predict gene expression in other samples, implying that the created patterns are largely generalizable. Interestingly, the impact of histone modifications on gene regulation appears predominantly indifferent to cellular specificity. Through explanation of the classifier's decisions, we substantiate established literature knowledge while concurrently revealing novel insights into the intricate landscape of transcriptional regulation via histone modification. The code for the PatternChrome algorithm, the scripts for the analyses and the required data can be found at (https://gitlab.gwdg.de/MedBioinf/generegulation/patternchrome). Supplementary data are available at Bioinformatics online.

Generative artificial intelligence enables the generation of bone scintigraphy images and improves generalization of deep learning models in data-constrained environments.

Advancements of deep learning in medical imaging are often constrained by the limited availability of large, annotated datasets, resulting in underperforming models when deployed under real-world conditions. This study investigated a generative artificial intelligence (AI) approach to create synthetic medical images taking the example of bone scintigraphy scans, to increase the data diversity of small-scale datasets for more effective model training and improved generalization. We trained a generative model on 99mTc-bone scintigraphy scans from 9,170 patients in one center to generate high-quality and fully anonymized annotated scans of patients representing two distinct disease patterns: abnormal uptake indicative of (i) bone metastases and (ii) cardiac uptake indicative of cardiac amyloidosis. A blinded reader study was performed to assess the clinical validity and quality of the generated data. We investigated the added value of the generated data by augmenting an independent small single-center dataset with synthetic data and by training a deep learning model to detect abnormal uptake in a downstream classification task. We tested this model on 7,472 scans from 6,448 patients across four external sites in a cross-tracer and cross-scanner setting and associated the resulting model predictions with clinical outcomes. The clinical value and high quality of the synthetic imaging data were confirmed by four readers, who were unable to distinguish synthetic scans from real scans (average accuracy: 0.48% [95% CI 0.46-0.51]), disagreeing in 239 (60%) of 400 cases (Fleiss' kappa: 0.18). Adding synthetic data to the training set improved model performance by a mean (± SD) of 33(± 10)% AUC (p < 0.0001) for detecting abnormal uptake indicative of bone metastases and by 5(± 4)% AUC (p < 0.0001) for detecting uptake indicative of cardiac amyloidosis across both internal and external testing cohorts, compared to models without synthetic training data. Patients with predicted abnormal uptake had adverse clinical outcomes (log-rank: p < 0.0001). Generative AI enables the targeted generation of bone scintigraphy images representing different clinical conditions. Our findings point to the potential of synthetic data to overcome challenges in data sharing and in developing reliable and prognostic deep learning models in data-limited environments.

Transformer-based neural speech decoding from surface and depth electrode signals

Objective.This study investigates speech decoding from neural signals captured by intracranial electrodes. Most prior works can only work with electrodes on a 2D grid (i.e. Electrocorticographic (ECoG) or ECoG array) and data from a single patient. We aim to design a deep-learning model architecture that can accommodate both surface ECoG and depth (stereotactic EEG or sEEG) electrodes. The architecture should allow training on data from multiple participants with large variability in electrode placements. The model should not have subject-specific layers and the trained model should perform well on participants unseen during training.Approach.We propose a novel transformer-based model architecture named SwinTW that can work with arbitrarily positioned electrodes by leveraging their 3D locations on the cortex rather than their positions on a 2D grid. We train subject-specific models using data from a single participant and multi-subject models exploiting data from multiple participants.Main results.The subject-specific models using only low-density 8 × 8 ECoG data achieved high decoding Pearson Correlation Coefficient with ground truth spectrogram (PCC = 0.817), overN= 43 participants, significantly outperforming our prior convolutional ResNet model and the 3D Swin transformer model. Incorporating additional strip, depth, and grid electrodes available in each participant (N= 39) led to further improvement (PCC = 0.838). For participants with only sEEG electrodes (N= 9), subject-specific models still enjoy comparable performance with an average PCC = 0.798. A single multi-subject model trained on ECoG data from 15 participants yielded comparable results (PCC = 0.837) as 15 models trained individually for these participants (PCC = 0.831). Furthermore, the multi-subject models achieved high performance on unseen participants, with an average PCC = 0.765 in leave-one-out cross-validation.Significance.The proposed SwinTW decoder enables future speech decoding approaches to utilize any electrode placement that is clinically optimal or feasible for a particular participant, including using only depth electrodes, which are more routinely implanted in chronic neurosurgical procedures. The success of the single multi-subject model when tested on participants within the training cohort demonstrates that the model architecture is capable of exploiting data from multiple participants with diverse electrode placements. The architecture's flexibility in training with both single-subject and multi-subject data, as well as grid and non-grid electrodes, ensures its broad applicability. Importantly, the generalizability of the multi-subject models in our study population suggests that a model trained using paired acoustic and neural data from multiple patients can potentially be applied to new patients with speech disability where acoustic-neural training data is not feasible.

The Social Construction of Categorical Data: Mixed Methods Approach to Assessing Data Features in Publicly Available Datasets.

In data-sparse areas such as health care, computer scientists aim to leverage as much available information as possible to increase the accuracy of their machine learning models' outputs. As a standard, categorical data, such as patients' gender, socioeconomic status, or skin color, are used to train models in fusion with other data types, such as medical images and text-based medical information. However, the effects of including categorical data features for model training in such data-scarce areas are underexamined, particularly regarding models intended to serve individuals equitably in a diverse population. This study aimed to explore categorical data's effects on machine learning model outputs, rooted the effects in the data collection and dataset publication processes, and proposed a mixed methods approach to examining datasets' data categories before using them for machine learning training. Against the theoretical background of thesocial construction of categories, we suggest a mixed methods approach to assess categorical data's utility for machine learning model training. As an example, we applied our approach to a Brazilian dermatological dataset (Dermatological and Surgical Assistance Program at the Federal University of Espírito Santo [PAD-UFES] 20). We first present an exploratory, quantitative study that assesses the effects when including or excluding each of the unique categorical data features of the PAD-UFES 20 dataset for training a transformer-based model using a data fusion algorithm. We then pair our quantitative analysis with a qualitative examination of the data categories based on interviews with the dataset authors. Our quantitative study suggests scattered effects of including categorical data for machine learning model training across predictive classes. Our qualitative analysis gives insights into how the categorical data were collected and why they were published, explaining some of the quantitative effects that we observed. Our findings highlight the social constructedness of categorical data in publicly available datasets, meaning that the data in a category heavily depend on both how these categories are defined by the dataset creators and the sociomedico context in which the data are collected. This reveals relevant limitations of using publicly available datasets in contexts different from those of the collection of their data. We caution against using data features of publicly available datasets without reflection on the social construction and context dependency of their categorical data features, particularly in data-sparse areas. We conclude that social scientific, context-dependent analysis of available data features using both quantitative and qualitative methods is helpful in judging the utility of categorical data for the population for which a model is intended.

PhysioEx, a new Python library for explainable sleep staging through deep learning.

&#xD;Sleep staging is a crucial task in clinical and research contexts for diagnosing and understanding sleep disorders. This work introduces PhysioEx, a Python library designed to support the analysis of sleep stages using deep learning and Explainable AI (XAI). &#xD;&#xD;Approach:&#xD;PhysioEx provides an extensible and modular API for standardizing and automating the sleep staging pipeline, covering data preprocessing, model training, testing, fine-tuning, and explainability. It supports both low-resource devices and high-performance computing clusters and includes pretrained models based on the Sleep Heart Health Study (SHHS) dataset. These models support single-channel EEG and multichannel EEG-EOG-EMG configurations and are easily adaptable to custom datasets. PhysioEx also features a command-line interface toolbox allowing users to streamline the model development and deployment. The library offers a range of XAI post-hoc methods to explain model decisions and align them with expert knowledge. &#xD;&#xD;Main results:&#xD;PhysioEx benchmark state-of-the-art sleep staging models in a standard pipeline. Enabling a fair comparison between them both on the training source and out-of-domain sources. Its XAI techniques provide insights into deep learning-based sleep staging by linking model decisions to human-understandable concepts, such as AASM-defined rules. &#xD;&#xD;Significance:&#xD;PhysioEx addresses the need for a standardized and accessible platform for sleep staging analysis, combining deep learning and XAI. By supporting modular workflows and explainable insights, it bridges the gap between machine learning models and clinical expertise. PhysioEx is publicly available and installable via pip, making it a valuable tool for researchers and practitioners in sleep medicine.

A Force-Sensor-Less Approach for Rapid Young's Modulus Identification of Heterogeneous Soft Tissue.

Due to individual differences, accurate identification of tissue elastic parameters is essential for biomechanical modeling in surgical guidance for hepatic venous injections. This paper aims to acquire the absolute Young's modulus of heterogeneous soft tissues during endoscopic surgery with 2D ultrasound images. First, we introduced a force-sensor-less approach that utilizes a pre-calibrated soft patch with a known Young's modulus and its ultrasound images to calculate the external forces exerted by the probe on the tissue. Second, we introduced a Kriging-based inverse algorithm to identify the relative Young's modulus (RYM) between the inclusion and the background tissue. The RYM was estimated based on 2D plane strain approximation and mapped to the RYM of 3D soft tissue through a trained Kriging model. Finally, we developed a direct method to identify the background Young's modulus (BYM) based on calculated external forces and RYM. The simulation results demonstrate the high efficiency and robustness of the Kriging-based inverse algorithm in identifying RYM. Physical experiments on the three phantoms show that the errors of the identified BYM and RYM are all below 15%. The proposed methodology for Young's modulus identification is feasible and achieves satisfactory accuracy and computational efficiency in both simulations and physical experiments.

Unified monitoring and telemetry platform supporting network intelligence in optical networks

In recent years, machine-learning (ML) applications have generated considerable interest and shown great potential in optimizing optical network management, such as quality of transmission estimation, traffic prediction, and resource allocation. However, these applications often require large datasets for training, inference, and updating, while network operators are generally reluctant to disclose their data due to privacy concerns and the sensitivity of operational information. Most open-source datasets typically lack transparency regarding network specifics, such as topology details and device configurations, making data acquisition and ML model training more difficult. In response, this paper presents a unified monitoring and telemetry platform that leverages distributed and centralized time-series databases on InfluxDB, a Kafka-based telemetry pipeline, and advanced ML applications. The separation of distributed and centralized databases improves data management flexibility and scalability. The Kafka-based telemetry pipeline ensures high-throughput, low-latency data streaming with end-to-end latency under 0.05 s through optimized partitioning. Additionally, integrating Kafka and InfluxDB allows for real-time data visualization from multiple sources, improving transparency and supporting real-time data streaming for network applications. By implementing this advanced telemetry and ML architecture, network operators can build a more intelligent, responsive, and resilient optical network infrastructure.

Intelligent classification of computer vulnerabilities and network security management system: Combining memristor neural network and improved TCNN model.

To enhance the intelligent classification of computer vulnerabilities and improve the efficiency and accuracy of network security management, this study delves into the application of a comprehensive classification system that integrates the Memristor Neural Network (MNN) and an improved Temporal Convolutional Neural Network (TCNN) in network security management. This system not only focuses on the precise classification of vulnerability data but also emphasizes its core role in strengthening the network security management framework. Firstly, the study designs and implements a neural network model based on memristors. The MNN, by simulating the memory effect of biological neurons, effectively captures the complex nonlinear relationships within vulnerability data, thereby enhancing the data insight capabilities of the network security management system. Subsequently, structural optimization and parameter adjustments are made to the TCNN model, incorporating residual connections and attention mechanisms to improve its classification performance, making it more adaptable to the dynamically changing network security environment. Through data preprocessing, feature extraction, and model training, this study conducts experimental validation on a public vulnerability dataset. The experimental results indicate that: The MNN model demonstrates excellent performance across evaluation metrics such as Accuracy (ACC), Precision (P), Recall (R), and F1 Score, achieving an ACC of 89.5%, P of 90.2%, R of 88.7%, and F1 of 89.4%. The improved TCNN model shows even more outstanding performance on the aforementioned evaluation metrics. After structural optimization and parameter adjustments, the TCNN model's ACC increases to 93.8%, significantly higher than the MNN model. The P value also improves, reaching 91.5%, indicating enhanced capability in reducing false positives and improving vulnerability identification accuracy. The integrated classification system, leveraging the strengths of both the MNN and improved TCNN models, achieves an ACC of 95.2%. This improvement not only demonstrates the system's superior capability in accurately classifying vulnerability data but also proves the synergistic effect of MNN and TCNN models in addressing complex network security environments. The comprehensive classification system proposed in this study significantly enhances the classification performance of computer vulnerabilities, providing robust technical support for network security management. The system exhibits higher accuracy and stability in handling complex vulnerability datasets, making it highly valuable for practical applications and research.

Using machine learning to forecast peak health care service demand in real-time during the 2022-23 winter season: A pilot in England, UK.

During winter months, there is increased pressure on health care systems in temperature climates due to seasonal increases in respiratory illnesses. Providing real-time short-term forecasts of the demand for health care services helps managers plan their services. During the Winter of 2022-23 we piloted a new forecasting pipeline, using existing surveillance indicators which are sensitive to increases in respiratory syncytial virus (RSV). Indicators including telehealth cough calls and emergency department (ED) bronchiolitis attendances, both in children under 5 years. We utilised machine learning techniques to train and select models that would best forecast the timing and intensity of peaks up to 28 days ahead. Forecast uncertainty was modelled usings a novel generalised additive model for location, scale and shape (gamlss) approach which enabled prediction intervals to vary according to the level of the forecast activity. The winter of 2022-23 was atypical because the demand for healthcare services in children was exceptionally high, due to RSV circulating in the community and increased concerns around invasive group A streptococcal (iGAS) infections. However, our short-term forecasts proved to be adaptive forecasting a new higher peak once the increasing demand due to iGAS started. Thus, we have demonstrated the utility of our approach, adding forecasts to existing surveillance systems.

Adversarial purification using a modified guided diffusion model and adaptive self-supervised training

Abstract As the applications of deep neural networks broaden, enhancing their robustness against adversarial attacks becomes increasingly important. Multiple studies have demonstrated their vulnerability to adversarial perturbations. This paper presents an iterative method employing a probabilistic diffusion model, guided by a self-supervised training strategy, to purify potential adversarial inputs. Specifically, the diffusion process blends the adversarial noise with incrementally added Gaussian noise. Subsequently, both types of noise are removed during the guided denoising process. Unlike existing methods that use a fixed number of iterations and rely on the adversarial input for guidance, our scoring-based approach dynamically adjusts the purification duration for each individual image, thereby reducing computational overhead and minimizing the side effects caused by excessive purification. Additionally, we modify the input at each iteration using a self-supervised strategy and then utilize this modified input to guide the denoising process, resulting in better adversarial robustness. Since the proposed method operates without any label information, it can be applied to various training paradigms, including supervised, semi-supervised, self-supervised, and unsupervised learning. We conduct several experiments on widely used machine vision datasets to evaluate the efficacy of the proposed method. The results confirm that our method effectively eliminates adversarial perturbations across these datasets. For example, under a white-box PGD attack with an l ∞ ball (ϵ = 8/255) on CIFAR-10, our method achieves a robust accuracy of 92.13%, surpassing the state-of-the-art by 3.68%.

Diabetes Prediction using Federated Learning

Federated learning is an innovative approach used in the medical field to resolve issues like centralization, privacy, and confidentiality. It gathers diverse data from several local models and aggregates it in a global model where only results are shared instead of data. This collaborative model training method aims for optimal performance. This study builds a framework for diabetes prediction using local models such as Artificial Neural Networks (ANN), Recurrent Neural Networks (RNN), and Long Short-Term Memory (LSTM) networks, trained independently on data distributed across multiple hospitals to ensure privacy and security. Techniques like Exploratory Data Analysis (EDA) and Synthetic Minority Over-sampling Technique (SMOTE) are used to improve datasets and address class imbalance. Among the models, ANN achieves the highest accuracy (89.99%), making it the preferred choice for predictions. This approach enhances the effectiveness and reliability of diabetes prediction systems in collaborative healthcare environments. Key Words: federated learning, diabetes prediction, privacy, artificial neural networks, exploratory data analysis, SMOTE

Deep Learning-Enhanced Portable Chemiluminescence Biosensor: 3D-Printed, Smartphone-Integrated Platform for Glucose Detection

A novel, portable chemiluminescence (CL) sensing platform powered by deep learning and smartphone integration has been developed for cost-effective and selective glucose detection. This platform features low-cost, wax-printed micro-pads (WPµ-pads) on paper-based substrates used to construct a miniaturized CL sensor. A 3D-printed black box serves as a compact WPµ-pad sensing chamber, replacing traditional bulky equipment, such as charge coupled device (CCD) cameras and optical sensors. Smartphone integration enables a seamless and user-friendly diagnostic experience, making this platform highly suitable for point-of-care (PoC) applications. Deep learning models significantly enhance the platform’s performance, offering superior accuracy and efficiency in CL image analysis. A dataset of 600 experimental CL images was utilized, out of which 80% were used for model training, with 20% of the images reserved for testing. Comparative analysis was conducted using multiple deep learning models, including Random Forest, the Support Vector Machine (SVM), InceptionV3, VGG16, and ResNet-50, to identify the optimal architecture for accurate glucose detection. The CL sensor demonstrates a linear detection range of 10–1000 µM, with a low detection limit of 8.68 µM. Extensive evaluations confirmed its stability, repeatability, and reliability under real-world conditions. This deep learning-powered platform not only improves the accuracy of analyte detection, but also democratizes access to advanced diagnostics through cost-effective and portable technology. This work paves the way for next-generation biosensing, offering transformative potential in healthcare and other domains requiring rapid and reliable analyte detection.

Privacy-Preserving Incentive Allocation for Fair and Resilient Data Sharing in Resource-Constrained Edge Computing Networks

Efficient and secure data sharing is paramount for advancing modern digital ecosystems, especially within edge computing environments characterized by resource-constrained nodes and dynamic network topologies. In such settings, privacy preservation, computational efficiency, and system resilience are critical for user engagement and overall system performance. However, existing approaches face three primary challenges: (i) limited optimization of privacy protection and absence of dynamic privacy budget scheduling for resource-constrained scenarios, (ii) static incentive mechanisms that overlook individual differences in data quality and resource consumption, and (iii) inadequate strategies to ensure resilience in environments with limited resources and unstable networks. This paper introduces the Federated Learning-based Dynamic Incentive Allocation Framework (FL-DIAF) to address these issues. FL-DIAF integrates differential privacy into the federated learning paradigm deployed on edge nodes, enabling collaborative model training that safeguards individual data privacy while maintaining computational efficiency and system resilience. Additionally, the framework employs a Shapley value-based dynamic incentive allocation model to ensure equitable and transparent distribution of incentives by accurately quantifying each participant’s contribution within an elastic edge computing infrastructure. Comprehensive experimental evaluations on diverse datasets demonstrate that FL-DIAF achieves a 9.573% reduction in the objective function value under typical conditions and attains a 100% task completion rate across all tested resilient edge scenarios.

Cataract Eye Disease Diagnosis Using the Random Forest Method

This study developed a machine learning-based classification model using the Random Forest algorithm to detect cataract risk based on 11 variables: age, gender, family history, lens opacity, visual acuity reduction, light sensitivity, color changes, double vision, intraocular pressure, slit-lamp results, and visual acuity. Feature importance analysis revealed that lens opacity and visual acuity variables contributed most significantly to cataract risk prediction, followed by intraocular pressure and visual acuity reduction. The system was designed using Google Colab for model training and Streamlit as an interactive interface, enabling real-time predictions with intuitive result visualization. After optimization using Grid Search, the model achieved an accuracy of 92.0%, precision of 95.0%, sensitivity of 90.0%, F1 Score of 92.4%, and specificity of 98.0%. This system is expected to serve as an effective supporting tool for medical professionals in the early diagnosis of cataracts.

Leveraging Incremental Machine Learning for Reconfigurable Systems Modeling under Dynamic Workloads

Dynamic workload orchestration is one of the main concerns when working with heterogeneous computing infrastructures in the edge-cloud continuum. In this context, FPGA-based computing nodes can take advantage of their improved flexibility, performance and energy efficiency provided that they use proper resource management strategies. In this regard, many state-of-the-art systems rely on proactive power management techniques and task scheduling decisions, which in turn require deep knowledge about the applications to be accelerated and the actual response of the target reconfigurable fabrics when executing them. While acquiring this knowledge at design time was more or less feasible in the past, with applications mostly being static task graphs that did not change at run time, the highly dynamic nature of current workloads in the edge-cloud continuum, where tasks can be deployed on any node and at any time, has removed this possibility. As a result, being able to derive such information at run time to make informed decisions has become a must. This paper presents an infrastructure to build incremental ML models that can be used to obtain run-time power consumption and performance estimations in FPGA-based reconfigurable multi-accelerator systems operating under dynamic workloads. The proposed infrastructure features a novel stop-and-restart resource-aware mechanism to monitor and control the model training and evaluation stages during normal system operation, enabling low-overhead updates in the models to account for either unexpected acceleration requests (i.e., tasks not considered previously by the models) or model drift (e.g., fabric degradation). Experimental results show that the proposed approach induces a maximal additional error of 3.66% compared to a continuous training alternative. Furthermore, the proposed approach incurs only a 4.49% execution time overhead, compared to the 20.91% overhead induced by the continuous training alternative. The proposed modeling strategy enables innovative scheduling approaches in reconfigurable systems. This is exemplified by the conflict-aware scheduler introduced in this work, which achieves up to a 1.35 times speedup in executing the experimental workload. Additionally, the proposed approach demonstrates superior adaptability compared to other methods in the literature, particularly in response to significant changes in workload and to mitigate the effects of model overfitting. The portability of the proposed modeling methodology and monitoring infrastructure is also shown through their application to both Zynq-7000 and Zynq UltraScale+ devices.

Assessment of Deep Neural Network Models for Direct and Recursive Multi-Step Prediction of PM10 in Southern Spain

Western Europe has been strongly affected in the last decades by Saharan dust incursions, causing a high PM10 concentration and red rain. In this study, dust events and the performance of seven neural network prediction models, including convolutional neural networks (CNN) and recurrent neural networks (RNN), have been analyzed in a PM10 concentration series from a monitoring station in Córdoba, southern Spain. The models were also assessed here for recursive multi-step prediction over different forecast periods in three different situations: background concentration, a strong dust event, and an extreme dust event. A very important increase in the number of dust events has been identified in the last few years. Results show that CNN models outperform the other models in terms of accuracy for direct 24 h prediction (RMSE values between 10.00 and 10.20 μg/m3), whereas the recursive prediction is only suitable for background concentration in the short term (for 2–5-day forecasts). The assessment and improvement of prediction models might help the development of early-warning systems for these events. From the authors’ perspective, the evaluation of trained models beyond the direct multi-step predictions allowed to fill a gap in this research field, which few articles have explored in depth.

Prediction of Individual Halo Concentrations Across Cosmic Time Using Neural Networks

The concentration of dark matter haloes is closely linked to their mass accretion history. We utilize the halo mass accretion histories from large cosmological N-body simulations as inputs for our neural networks, which we train to predict the concentration of individual haloes at a given redshift. The trained model performs effectively in other cosmological simulations, achieving the root mean square error between the actual and predicted concentrations that significantly lower than that of the model by Zhao et al. and Giocoli et al. at any redshift. This model serves as a valuable tool for rapidly predicting halo concentrations at specified redshifts in large cosmological simulations.

Mapping National‐Scale Road Surface Types Using Multisource Open Data and Deep Learning Model

ABSTRACTIdentifying road surface type (e.g., paved and unpaved) is crucial for pavement maintenance, transportation management, and road network accessibility research. Existing approaches relying on vehicle‐mounted devices or remote sensing data are have limitation for large‐scale road networks. This study proposes a novel approach to identify national‐scale road surface type using multiple open geospatial datasets and machine learning models. Specifically, 16 input variables were designed based on these datasets (including OpenStreetMap, GDP, population, building height, and land cover). Nigeria and Cameroon were selected as study areas. A substantial dataset, auto‐extracting road surface tags from OpenSreetMap, was used to train a model. The trained model predicted road surface types across the two study areas. Result indicated: (1) Most of the input variables positively impact the output variable, with “road class” being the most influential; (2) The proposed approach with deep learning model‐TabNet performs the best, with an overall accuracy above 85%; and (3) More than 83% of roads in the two African countries are unpaved, with paved roads concentrated in backbone roads and southern provinces. This approach has been validated and offers valuable insights for local authorities aiming to enhance road infrastructure.

Comprehensive exploration of diffusion models in image generation: a survey

The rapid development of deep learning technology has led to the emergence of diffusion models as a promising generative model with diverse applications. These include image generation, audio and video synthesis, molecular design, and text generation. The distinctive generation mechanism and exceptional generation quality of diffusion models have made them a valuable tool in these diverse fields. However, with the extensive deployment of diffusion models in the domain of image generation, concerns pertaining to data privacy, data security, and artistic ethics have emerged with increasing prominence. Given the accelerated pace of development in the field of diffusion models, the majority of extant surveys are deficient in two respects: firstly, they fail to encompass the latest advances in diffusion-based image synthesis; and secondly, they seldom consider the potential social implications of diffusion models. In order to address these issues, this paper presents a comprehensive survey of the most recent applications of diffusion models in the field of image generation. Furthermore, it provides an in-depth analysis of the potential social impacts that may result from their use. Firstly, this paper presents a systematic survey of the background principles and theoretical foundations of diffusion models. Subsequently, this paper provides a detailed examination of the most recent applications of diffusion models across a range of image generation subfields, including style transfer, image completion, image editing, super-resolution, and beyond. Finally, we present a comprehensive examination of these social issues, addressing data privacy concerns, such as the potential for data leakage and the implementation of protective measures during model training. We also analyse the risk of malicious exploitation of the model and the defensive strategies employed to mitigate such risks. Additionally, we examine the implications of the authenticity and originality of generated images on artistic creativity and copyright protection.