Computational Resources Research Articles

Evaluating Text-to-Image Generation Methods: Stable Diffusion vs Generative Adversarial Networks (GANs)

Text-to-image generation is a fast developing field of artificial intelligence that allows verbal descriptions to be converted into realistic or creative visuals. This study investigates the differences between two cutting-edge approaches for generating images from text, examining their performance, efficiency, and practical applicability across multiple areas. The dominant techniques in this discipline are Generative Adversarial Networks (GANs) and Stable Diffusion models. While GANs have long been the preferred architecture for picture generation tasks, newer diffusion-based models such as Stable Diffusion have emerged as viable alternatives, providing distinct methods to noise reduction and image synthesis. Attention GAN(AttnGAN), a GAN-based approach, uses attention mechanisms to improve the semantic alignment of text descriptions and generated images, resulting in more contextually appropriate graphics. These methodologies are compared, with an emphasis on architectural differences, performance, and applicability to varied applications. GANs use adversarial training, in which two networks (the generator and the discriminator) compete to produce increasingly realistic images. This method is quite effective for producing high-quality photos, but it has drawbacks such as mode collapse and training instability. In contrast, Stable Diffusion models use a probabilistic diffusion process to iteratively reduce noisy images into coherent outputs, resulting in increased processing efficiency and the ability to handle high-resolution images. Experimental evaluation of benchmark datasets reveals each method's strengths and limits in real applications such as digital art, content development, and product design. Stable Diffusion produces more diverse and high-resolution images with fewer computer resources, but GANs generate extremely detailed and realistic visuals. The comparative insights gathered from this research can be used to choose the best technique for a given text-to-image production problem

Deep Metric Learning with Augmented Latent Fusion and Response-Based Knowledge Distillation on Edge Device for Paddy Pests and Disease Identification

The health of paddy fields significantly impacts rice yields and the economic stability of farmers. Limited number of experts available to watch these issues poses a challenge. Consequently, a reliable diagnostic system is necessary to find pests and diseases in rice crops. In this study, we propose deep metric learning with augmented latent fusion (FADMAKA) combined with a response-based knowledge distillation (KD) approach. The student model, which processes single RGB input images, is trained using soft latent labels derived from four augmented input from the teacher model. Our method delivers a high validation accuracy of 0.973, keeps an accuracy of 0.782 on the unseen data, and with rapid inference time of 38.911 milliseconds. This approach’s accuracy outperforms SoftMax deep learning classification with fine-tuning, which only has a maximum accuracy of 0.739 on the unseen data with computation time of 36.224 ms, and the DML with augmented latent fusion with k-NN classifier on the same base model, which achieves an accuracy of 0.78 with computation time of 124.977 ms. Our proposed model has 0.12 giga floating point operations per second (GFLOPs) that is suitable for edge devices with low computational resources. Following the modeling phase, we deployed the highest-accuracy student model to a Raspberry Pi 4B device equipped with a camera. This system can provide biological agent-based recommendations for identified pest and disease threats in rice fields. Our approach not only improved accuracy but also proved efficiency, enabling farmers to identify pests and disease without relying on internet connectivity.

Accelerating ab initio melting property calculations with machine learning: application to the high entropy alloy TaVCrW

Melting properties are critical for designing novel materials, especially for discovering high-performance, high-melting refractory materials. Experimental measurements of these properties are extremely challenging due to their high melting temperatures. Complementary theoretical predictions are, therefore, indispensable. One of the most accurate approaches for this purpose is the ab initio free-energy approach based on density functional theory (DFT). However, it generally involves expensive thermodynamic integration using ab initio molecular dynamic simulations. The high computational cost makes high-throughput calculations infeasible. Here, we propose a highly efficient DFT-based method aided by a specially designed machine learning potential. As the machine learning potential can closely reproduce the ab initio phase-space distribution, even for multi-component alloys, the costly thermodynamic integration can be fully substituted with more efficient free energy perturbation calculations. The method achieves overall savings of computational resources by 80% compared to current alternatives. We apply the method to the high-entropy alloy TaVCrW and calculate its melting properties, including the melting temperature, entropy and enthalpy of fusion, and volume change at the melting point. Additionally, the heat capacities of solid and liquid TaVCrW are calculated. The results agree reasonably with the CALPHAD extrapolated values.

A modified and extended genetic algorithm for optimal distributed generation grid-integration solutions in direct current power grids

The integration of distributed generation into direct current power grids presents a critical challenge in modern energy systems, as it directly impacts grid reliability, efficiency, and the successful transition to renewable energy. This study addresses the problem of optimizing distributed generation placement and sizing in direct current grids, a key issue for reducing power losses and improving energy distribution. To tackle this, a modified and extended genetic algorithm was developed, capable of handling both continuous and discrete variables simultaneously. The algorithm was tested on two standard direct current grid systems, a 21-bus microgrid and a 69-bus network. The results demonstrated significant improvements over existing methods, reducing power losses by 84.5% in the 21-bus microgrid and by 95% in the 69-bus direct current network, with notably reduced computation times. These findings indicate that the proposed algorithm not only optimizes distributed generation integration effectively but also offers superior performance compared to traditional approaches, without the need for additional methods or software. The novelty of this work lies in its ability to handle complex, nonlinear optimization problems within direct current grids using a single, efficient approach, advancing beyond previous efforts by achieving better results with fewer computational resources.

Automated MRI quantification of pediatric abdominal adipose tissue using convolutional neural networks and novel total intensity maps

ObjectiveChildhood obesity is a significant global health concern. Visceral adipose tissue (VAT) is more closely linked to the development of metabolic and cardiovascular diseases than abdominal subcutaneous adipose tissue (ASAT). This study aimed to develop a straightforward and automated method to quantify both ASAT and VAT in children from MRI. MethodsDixon-based 3 T-MRI scans were conducted on Mexican boys aged 7–9 years. Novel Total Intensity Maps were generated from these scans for input into simple 2D Convolutional Neural Networks (CNNs) for automatic prediction/quantification of ASAT and VAT. The results were compared with the commercial semi-automated quantification and validated method provided by AMRA® Researcher which is the gold standard. ResultsThe CNN-based method produced ASAT and VAT quantifications comparable to those obtained by AMRA® Researcher (ASAT, p = 0.1765; VAT, p = 0.7757), with a high correlation between the two methods for ASAT (R2 = 0.98, p < 10−14) and VAT (R2 = 0.95, p < 10−11). The reproducibility coefficients (RPC) were 0.32 L for ASAT and 0.17 L for VAT with our novel methodology. ConclusionThis study presents a simple automated method for quantifying ASAT and VAT in children from MRI data using 2D-CNN analysis of novel inputs, Total Intensity Maps. The new method produced measurements comparable to those obtained by AMRA® Researcher on the same dataset. SignificanceThe proposed automatic and straightforward quantification method, with minimal computational resource requirements, may serve as an accessible tool for researchers and clinicians in the diagnosis, follow-up, and prevention of childhood obesity.

Joint edge caching and computation offloading for heterogeneous tasks in MEC-enabled vehicular networks

Edge caching is an effective paradigm that can significantly reduce the computation task offloading latency for mobile edge computing (MEC) in vehicular networks, while also alleviating the backhaul transmission pressure for retrieving content data from the cloud server. However, most existing works fail to address how to handle heterogeneous tasks generated by vehicle terminals (VTs), especially in complex scenarios where both computation and content tasks are generated simultaneously. In this paper, we consider a mobility-aware vehicular network model where VTs simultaneously generate heterogeneous task requests, i.e., a computation task and a content task, and investigate joint optimization of caching for heterogeneous tasks data, computation offloading, and computing resource allocation. In order to optimize the latency for processing the heterogeneous tasks, an average execution latency minimization problem with sojourn time and caching capacity constraints is formulated. We decompose this problem into two tractable subproblems, i.e., caching optimization subproblem, and computation offloading and resource allocation optimization subproblem. We first develop a dynamic programming (DP) algorithm to obtain the optimal caching strategies for heterogeneous tasks data. We compare the obtained content retrieval latency with the local computing latency, and derive the optimal computation offloading and edge computing resource allocation solutions. On this basis, we propose a joint computation offloading and resource allocation (JCORA) algorithm to determine the computing resources allocated to each VT and corresponding computation offloading strategy. Numerical results indicate that the proposed algorithm, which integrates DP algorithm and JCORA algorithm, can achieve lower execution latency for heterogeneous tasks compared to the benchmark schemes. Additionally, for task loss scenarios where the sojourn time constraint cannot be met, the impact of VT mobility on the task loss probability is also revealed.

Land surface scheme TerM: the model formulation, code architecture and applications

Abstract This paper presents the INM RAS–MSU land surface scheme, extracted from the INM RAS Earth system model into an independent software complex and supplemented with several modules to reproduce new components and processes of the Earth system. The resulting software product is referred to as TerM (Terrestrial Model). The physical and mathematical foundations of the model, the main features of the software implementation, and examples of applications in reproducing components of the terrestrial hydrological and carbon cycles are briefly outlined. Separating the land surface block into a standalone software complex significantly saves computational resources when assessing the impact of global and regional climate changes on natural resources (including hydrological ones), ecosystem dynamics, and emissions of climate-relevant substances with high spatial detalization. Within the TerM modelling complex, the development, validation, and calibration of new parameterizations of physical and biogeochemical processes are being conducted in an autonomous mode for subsequent implementation into the full INM RAS Earth system model.

Effect of statistical positional correlation on the radiative property investigation of dispersed particulate medium

Spectral radiative transfer between particles of dispersed particulate medium is prevalent in various fields, which may be affected by particle distribution when considering dependent scattering effect (DSE). Meanwhile, statistical positional correlation (SPC) describes the possible existing order of particle distribution in the spatial variation for disordered dispersed particulate medium. SPC plays a pivotal role in the transformation of radiative transfer regimes between particles. However, the existing theory lacks a precise criterion for describing SPC and fails to comprehensively consider key factors within SPC influencing radiative transfer regimes, such as mean distance, relative distance, standard deviation, cluster, and aggregation. To achieve more representative and accurate results of radiative property calculations while minimizing computational resources, we proposed MRSDL (Mean distance, Relative distance, Standard deviation, Density-based clustering, and Line matrix) criterion combined with particle swarm optimization (PSO) for characterizing and regulating SPC. Moreover, to further achieve the statistical averaging of radiative properties precisely, the multiple sphere T-matrix (MSTM) method is combined to eliminate the random fluctuations of radiative properties caused by SPC. Compared to the conventional method, the method by authors can decrease error between experimental and calculation data from 41.23 % to 5.32 %, when considering the effect of SPC on the radiative property.

EARLY DETECTION OF BRAIN TUMOR USING MRI AND TRANSFER LEARNING

Brain tumors pose significant risks to cognitive functions, making early detection crucial for improving patient survival rates. Accurate tumor detection aids neuro-oncologists in diagnosing tumor types and recommending appropriate treatments. However, manual detection is challenging, time-consuming, and prone to human error. Recently, Deep Learning (DL) models have demonstrated substantial potential in efficiently classifying large image datasets. This paper presents three novel and efficient multi-class DL architectures leveraging transfer learning to classify different brain tumors. Our primary contribution is to enhance the predictive accuracy while minimizing the usage of computational resource compared to other state-of-art models in the literature by forgoing preprocessing, segmentation, hybrid models, and augmentation techniques. Additionally, we reduce the number of FC layers to streamline computation. We conduct extensive experiments to evaluate the performance of our models using the brain tumor Figshare T1-weighted contrast-enhanced MRI dataset, comprising 3064 images from three distinct tumor types. The InceptionV3 model records a 98.36% accuracy level with five-fold cross-validation. By incorporating batch normalization and optimizing the learning rate for the Adam optimizer, the Xception model reached 99.19% accuracy. Finally, we utilize Particle Swarm Optimization (PSO) to fine-tune the learning rate of the Stochastic Gradient Descent (SGD) optimizer. The Xception model attained 98.85% accuracy. These results highlight the novelty of our approach, offering a practical solution for neuro-oncologists, particularly through the fine-tuned Xception model, which delivers early and accurate tumor detection with minimal computational resources.

MGL-YOLO: A Lightweight Barcode Target Detection Algorithm

Due to the critical importance of one-dimensional barcode detection in logistics, retail, and manufacturing, which has become a key issue affecting operational efficiency, researchers have shown increasing interest in this area. However, deploying deep convolutional neural networks on embedded and some edge devices is very challenging due to limited storage space and computational resources. To address this issue, this paper proposes MGL-YOLO, a lightweight one-dimensional barcode detection network based on an improved YOLOv8, which aims to achieve a high detection accuracy at low computational cost. First, a new multi-scale group convolution (MSGConv) is designed and integrated into the C2f module to construct the MSG-C2f feature extraction module. By replacing the C2f module in the P5 layer of the backbone network, the ability to extract multi-scale feature information is enhanced. Secondly, a feature extraction module, Group RepConv Cross Stage Partial Efficient Long-Range Attention Network (GRCE), is designed to optimize the feature extraction capability of the C2f modules in the neck section, offering significant advantages in multi-scale characteristics and complexity adjustment. Finally, a Lightweight Shared Multi-Scale Detection Head (LSMD) is proposed, which improves the model’s detection accuracy and adaptability while reducing the model’s parameter size and computational complexity. Experimental results show that the proposed algorithm increases MAP50 and MAP50.95 by 2.57% and 2.31%, respectively, compared to YOLOv8, while reducing parameter size and computational cost by 36.21% and 34.15%, respectively. Moreover, it also demonstrates advantages in average precision compared to other object detection networks, proving the effectiveness of MGL-YOLO for one-dimensional barcode detection in complex backgrounds.

Resource Sizing for Virtual Environments of Networked Interconnected System Services

Networked interconnected systems are often deployed in infrastructures with resource allocation using isolated virtual environments. The technological implementation of such systems varies significantly, making it difficult to accurately estimate the required volume of resources to allocate for each virtual environment. This leads to overprovisioning of some services and underprovisioning of others. The problem of distributing the available computational resources between the system services arises. To efficiently use resources and reduce resource waste, the problem of minimizing free resources under conditions of unknown ratios of resource distribution between services is formalized; an approach to determining regression dependencies of computing resource consumption by services on the number of requests and a procedure for efficient resource distribution between services are proposed. The proposed solution is experimentally evaluated using the networked interconnected system model. The results show an increase in throughput by 20.75% compared to arbitrary resource distribution and a reduction in wasted resources by 55.59%. The dependences of the use of resources by networked interconnected system services on the number of incoming requests, identified using the proposed solution, can also be used for scaling in the event of an increase in the total volume of allocated resources.

Wind Turbine Wake Regulation Method Coupling Actuator Model and Engineering Wake Model

The wake effect is one of the main factors affecting the power generation of wind farms. Wake regulation is often used to reduce the wake interference between wind turbines. Accurate assessment of the wake flow of wind turbine is essential to wake regulation. Engineering wake models are widely used for rapid evaluation of the wake at present due to lower computational resource cost. However, the selection of empirical parameters of the wake model has significant influence on the prediction accuracy, especially in the case of yaw. The actuator model based on CFD simulation has less dependence on empirical parameters and higher simulation accuracy. However, the computational cost is too high for wake regulation for large wind farms. This paper proposed an improved wake regulation method that combines the advantages of the actuator line model (ALM) method and the engineering wake mode. The simulation results of the ALM is used to calibrate the empirical parameters of the engineering wake model. The calibrated wake model can be used to optimize the yaw angle of wind turbines during wake regulation. The accuracy of two models is compared using wind tunnel experimental data. The ALM results give better agreement to the experimental data. The Horns Rev wind farm case is used for the coupled method verification. The power generation increase using the engineering wake model is obviously greater than that of the ALM. After calibrating the wake model, the gap between the two power predictions is greatly narrowed, which proves the effectiveness of the proposed method. The proposed coupling method can be used to improve the credibility of the wake regulation with affordable computational cost.

An automatic IoT application placement using an improved equilibrium optimizer algorithm in fog computing

PurposeThe rapid proliferation of Internet of Things (IoT) devices across various domains has created a demand for real-time computing resources that traditional cloud computing models struggle to meet. Fog computing, which brings computation resources closer to IoT devices, has emerged as a promising solution. An automatic service placement framework is needed to use fog computing resources efficiently.Design/methodology/approachIn this study, first a three-layer independent service framework is introduced to define relationships between IoT devices and fog layers, facilitating automatic application deployment. Next, an enhanced version of the equilibrium optimizer (EO) algorithm, inspired by physics, is designed for service placement in fog computing environments.FindingsSimulations reveal that the proposed approach surpasses existing methods, achieving a 99% success rate compared to the closest alternative’s 93%. The algorithm also significantly reduces waiting and planning times for service placement, proving its efficiency and effectiveness in optimizing IoT service deployment in fog computing.Research limitations/implicationsOne of the primary limitations is the computational complexity involved in dynamically adjusting to real-time changes in network conditions and IoT workloads. Although improved EO offers improvements in placement efficiency, it may not be fully optimized for highly fluctuating environments. Another important limitation is the uncertainty in node resources. Fog computing environments often face unpredictable changes in the availability and capacity of resources across nodes. This uncertainty can affect the algorithm’s ability to consistently make optimal decisions for IoT service placement.Practical implicationsFrom a practical perspective, the implementation of the proposed framework and the improved EO algorithm can drastically enhance the efficiency of IoT service deployment in fog computing systems. Organizations that rely on IoT networks, particularly those with critical real-time requirements, can benefit from reduced service placement times and lower failure rates. This can lead to better resource utilization, reduced operational costs and improved overall performance of IoT systems. The commercial impact is evident in industries such as smart cities, healthcare, where fast data processing is crucial.Social implicationsOur proposed framework has important implications for real-world IoT applications, particularly in areas requiring low latency processing, such as healthcare, smart cities. By reducing service delays and optimizing resource allocation, the framework can significantly improve the quality and reliability of services. Additionally, improved resource management leads to cost savings and better system efficiency, making the technology accessible to a wider range of applications.Originality/valueExisting resource placement strategies have shown inadequate performance, highlighting the need for more advanced algorithms. This study introduces a three-layer automatic framework for enhancing the application deployment of a fog system beside a novel improved EO algorithm to offer a robust solution for assigning IoT applications to fog nodes.

A Review of Research on Object Detection Algorithms

Object detection is a fundamental task in computer vision, widely used in fields such as autonomous driving, security surveillance, medical imaging, and drone image analysis. With the continuous advancement of technology, object detection algorithms have evolved from traditional methods to deep learning approaches. This paper categorizes object detection algorithms into four types based on their technical characteristics and implementation methods: two-stage algorithms, one-stage algorithms, keypoint-based algorithms, and emerging Transformer-based methods. Through a performance comparison on existing datasets, it was found that two-stage algorithms excel in accuracy but consume significant computational resources, leading to slower speeds; one-stage algorithms have a clear advantage in speed but show lower accuracy in detecting small objects; keypoint-based methods effectively balance speed and accuracy; additionally, the emerging Transformer-based methods perform well in capturing global information but require large amounts of training data and computational resources. This paper summarizes the advantages and disadvantages of each type of algorithm and discusses future research directions.

Physical Reservoir Computing in a Music Hall Experiment

Abstract Music is a complex vibratory structure that evolves temporally and, while it has been studied for centuries, both quantitatively and qualitatively, it has seldom been studied under the lens of computer science and information theory. Further, while much research has been devoted to measuring and optimizing the acoustics of music venues, the computational ability of these spaces has never been explored. Using physical reservoir computing, this paper presents an experimental verification that a music hall has computational ability. Two experimental setups are explored: one has a single speaker and information is sent sequentially, and another has two speakers and information is sent simultaneously. Both of these exhibit qualitatively similar results. Thus, music might be, at least in part, a computational experience. The findings of this paper could provide a computational basis for the upper limit of tactus in music. To the authors' knowledge, this is the first time that a music hall has been utilized as a computational resource. Moreover, the computational ability of musical structures provides another tool to understand the complex relationship between music, vibrations, and human perception.

Application of Stable Diffusion and LoRA Models in AI Drawing

Abstract. This paper explores the application of Stable Diffusion model and LoRA (Low-Rank Adaptation) model in AI-generated artwork. The authors introduce the foundational principles of Stable Diffusion model and LoRA, as well as their application in high-quality image generation. Using three popular datasets ImageNet, COCO, and CelebA we apply various image quality assessment metrics, including PSNR (Peak Signal-to-Noise Ratio), IS (Inception Score), and FID (Frchet Inception Distance), and further validate their potential in artistic creation through subjective evaluations. By comparing the performance of these two models across different datasets, we examine their strengths, weaknesses, areas for improvement in image generation tasks, along with user experience considerations. The experimental results show that the Stable Diffusion model excels in terms of image quality and diversity, while the LoRA model offers significant benefits in computational efficiency and resource usage. Through comprehensive experimental evaluations, this paper provides a scientific basis for model selection in AI art creation and offers insights for the future development of hybrid models.

Advancements and Prospects in Large Integer Factorization: A Comprehensive Review of the Number Field Sieve Method

The Number Field Sieve (NFS) stands as one of the most effective algorithms for the factorization of large integers, playing a critical role in cryptographic security, particularly in breaking RSA encryption systems. This review paper provides an in-depth analysis of NFS, highlighting its structure, key processes, and computational challenges. NFS is notably effective for integers larger than 110 digits and is recognized for its sub-exponential computational complexity, making it superior to other classical factorization algorithms. The core processes of NFSpolynomial selection, number pair sieving, matrix solving, and square root extractionare systematically examined to illustrate their roles in factorization. This study also reviews the latest advancements in NFS, including improvements in polynomial selection methods and optimizations in sieving and matrix-solving stages, which have significantly enhanced the algorithms efficiency. Moreover, the paper discusses the future prospects of NFS, emphasizing the need for further optimization to reduce computational resource demands and increase practicality in large-scale applications. The ongoing evolution of NFS continues to push the boundaries of cryptographic analysis, driving future research towards even more efficient factorization methods.

An Efficient Multi-Level 2D DWT Architecture for Parallel Tile Block Processing with Integrated Quantization Modules

A multi-level 2D Discrete wavelet transform (DWT) architecture for JPEG2000 is proposed, enhancing speed through parallel processing multiple tile blocks. Based on the lifting scheme, folded architecture and unfolded architecture achieving critical path delay with only one multiplier are designed to increase throughput rate. Connecting the folded and unfolded architecture through a pipeline architecture ensures uniform throughput rates across all DWT levels within a singular clock domain. Computational resource consumption is reduced by adjusting the timing to allow one folded architecture to process three tile blocks of three to five levels of DWT, and a transposing module requiring merely six registers is devised to decrease storage resource consumption. The quantization module, crucial for code-word control in JPEG2000, is integrated into the scaling module with minimal additional resource expenditure. Compared to the existing architecture, the analysis demonstrates that the proposed architecture exhibits enhanced hardware efficiency, with a reduction in transistor-delay-product (TDP) of no less than 14.69%. Synthesis results further reveal an area reduction of at least 26.64%, and a decrease in area-delay-product (ADP) by a minimum of 29.89%. Results from FPGA implementation indicate a significant decrease in resource utilization.

Progress and Application of Unsupervised Feature Extraction Methods

Unsupervised feature extraction is crucial in machine learning and data mining for handling high-dimensional and unlabeled data. However, existing methods often ignore feature relationaships, resulting in suboptimal feature subsets. This paper reviews the current state of unsupervised feature extraction methods, discussing the limitations of traditional methods such as Principal Component Analysis (PCA) and Independent Component Analysis (ICA), particularly in terms of interpretability, sensitivity to outliers, and computational resource challenges. In recent years, improvement strategies such as information theory, sparse learning, and deep learning (e.g., deep autoencoders and generative adversarial networks) have significantly progressed in feature extraction. This paper analyzes the practical applications of these methods in image processing, gene analysis, text mining, and network security. For example, in image processing, deep autoencoder-based methods such as Matrix Capsules with EM Routing can effectively extract key features from complex images. In text mining, unsupervised feature selection methods combined with generative adversarial networks significantly improve the efficiency of processing high-dimensional text data. Additionally, this paper explores future research directions such as multimodal data processing, improving real-time processing capabilities, and integration with other machine learning techniques (e.g., reinforcement learning, transfer learning), providing insights for further development of unsupervised feature extraction technologies.

REDIportal: towardan integrated view of the A-to-I editing.

A-to-I RNA editing is the most common non-transient epitranscriptome modification. It plays several roles in human physiology and has been linked to several disorders. Large-scale deep transcriptome sequencing has fostered the characterization of A-to-I editing at the single nucleotide level and the development of dedicated computational resources. REDIportal is a unique and specialized database collecting ∼16 millionof putative A-to-I editing sites designed to face the current challenges of epitranscriptomics. Its running version has been enriched with sites from the TCGA project (using data from 31 studies). REDIportal provides an accurate, sustainable and accessible tool enriched with interconnections with widespread ELIXIR core resources such as Ensembl, RNAcentral, UniProtand PRIDE. Additionally, REDIportal now includes information regarding RNA editing in putative double-stranded RNAs, relevant for the immune-related roles of editing, as well as an extended catalog of recoding events. Finally, we report a reliability score per site calculated using a deep learning model trained using a huge collection of positive and negative instances. REDIportal is available at http://srv00.recas.ba.infn.it/atlas/.