Architecture Model Research Articles

Targeted generative data augmentation for automatic metastases detection from free-text radiology reports

Automatic identification of metastatic sites in cancer patients from electronic health records is a challenging yet crucial task with significant implications for diagnosis and treatment. In this study, we demonstrate how advancements in natural language processing, namely the instruction-following capability of recent large language models and extensive model pretraining, made it possible to automate metastases detection from radiology reports texts with a limited amount of gold-labeled data. Specifically, we prompt Llama3, an open-source instruction-tuned large language model, to generate synthetic training data to expand our limited labeled data and adapt BERT, a small pretrained language model, to the task. We further investigate three targeted data augmentation techniques which selectively expand the original training samples, leading to comparable or superior performance compared to vanilla data augmentation, in most cases, while being substantially more computationally efficient. In our experiments, data augmentation improved the average F1-score by 2.3, 3.5, and 3.9 points for lung, liver, and adrenal glands, the organs for which we had access to expert-annotated data. This observation suggests that Llama3, which has not been specifically tailored to this task or clinical data in general, can generate high-quality synthetic data through paraphrasing in the clinical context. We also compare metastasis identification accuracy between models utilizing institutionally standardized reports vs. non-structured reports, which complicate the extraction of relevant information, and show how including patient history with a customized model architecture narrows the gap between those two setups from 7.3 to 4.5 points on F1-score under LoRA tuning. Our work delivers a broadly applicable solution with remarkable performance that does not require model customization for each institution, making large-scale, low-cost spatio-temporal cancer progression pattern extraction possible.

When Cybersecurity Meets Accessibility: A Holistic Development Architecture for Inclusive Cyber-Secure Web Applications and Websites

In today’s technology-centric era, digital transformation has made accessibility and cybersecurity critical components for developing inclusive web applications that are essential in daily operations. However, current development models often treat accessibility and cybersecurity separately, leading to solutions that may excel in one area while lacking in the other. This paper addresses this gap by proposing the Holistic Security and Accessibility Layered Architecture (HSALA), a novel architectural model that integrates accessibility and cybersecurity throughout the development process. HSALA consists of five interconnected layers, each paired with a corresponding security layer, ensuring that both accessibility and security are embedded at every level. We conduct a theoretical and analytical case study comparing HSALA with the traditional three-layered architecture, demonstrating HSALA’s superiority in addressing contemporary challenges related to accessibility, cybersecurity, scalability, and user experience. The findings indicate that HSALA provides a robust framework for developing secure and inclusive web applications, promoting a more equitable digital ecosystem. We outline a roadmap for future work, including empirical validation and practical implementation strategies. HSALA represents a significant advancement in web application development, fostering inclusivity and trust in an increasingly digital society.

Quantifying and Exploiting Adversarial Vulnerability: Gradient-Based Input Pre-Filtering for Enhanced Performance in Black-Box Attacks

We investigate the vulnerability of inputs in an adversarial setting and demonstrate that certain samples are more susceptible to adversarial perturbations compared to others. Specifically, we employ a simple yet effective approach to quantify the adversarial vulnerability of inputs, which relies on the clipped gradients of the loss with respect to the input. Our observations indicate that inputs with a low percentage of zero gradient components tend to be more vulnerable to attacks. These findings are supported by a theoretical explanation on a linear model and empirical evidence on deep neural networks. Across all datasets we tested, we find that inputs with the lowest zero gradient percentage, on average, exhibit 34.5% more susceptibility to adversarial attacks than randomly selected inputs. Additionally, we demonstrate that the zero gradient percentage, as a metric, transfers across different model architectures. Finally, we propose a novel black-box attack pipeline that enhances the efficiency of conventional query-based black-box attacks and show that input pre-filtering based on ZGP can boost the attack success rates, particularly under low perturbation levels. On average, across all datasets we test, our approach outperforms the conventional shadow model-based and query-based black-box attack pipelines by 44.9% and 30.4%, respectively.

Deep learning in gonarthrosis classification: a comparative study of model architectures and single vs. multi-model methods

PurposeThis study aims to classify Kellgren–Lawrence (KL) osteoarthritis stages using knee anteroposterior X-ray images by comparing two deep learning (DL) methodologies: a traditional single-model approach and a proposed multi-model approach. We addressed three core research questions in this study: (1) How effective are single-model and multi-model deep learning approaches in classifying KL stages? (2) How do seven convolutional neural network (CNN) architectures perform across four distinct deep learning tasks? (3) What is the impact of CLAHE (Contrast Limited Adaptive Histogram Equalization) on classification performance?ApproachWe created a dataset of 14,607 annotated knee AP X-rays from three hospitals. The knee joint region was isolated using a YOLOv5 object detection model. The multi-model approach utilized three DL models: one for osteophyte detection, another for joint space narrowing analysis, and a third to combine these outputs with demographic and image data for KL classification. The single-model approach directly classified KL stages as a benchmark. Seven CNN architectures (NfNet-F0/F1, EfficientNet-B0/B3, Inception-ResNet-v2, VGG16) were trained with and without CLAHE augmentation.ResultsThe single-model approach achieved an F1-score of 0.763 and accuracy of 0.767, outperforming the multi-model strategy, which scored 0.736 and 0.740. Different models performed best across tasks, underscoring the need for task-specific architecture selection. CLAHE negatively impacted most models, with only one showing a marginal improvement of 0.3%.ConclusionThe single-model approach was more effective for KL grading, surpassing metrics in existing literature. These findings emphasize the importance of task-specific architectures and preprocessing. Future studies should explore ensemble modeling, advanced augmentations, and clinical validation to enhance applicability.

Screening of multi deep learning-based de novo molecular generation models and their application for specific target molecular generation

Traditional virtual screening methods need to explore expanse and vast chemical spaces and need to be based on existing chemical libraries. With the development of deep learning techniques for the de novo generation of molecules, also known as inverse molecular design, the increasingly widespread application of various types of deep learning algorithms has led to revolutionary changes in de novo molecular generation research. In particular, the emergence of a novel natural language processing (NLP) architecture called the transformer has improved the state-of-the-art performance of existing AI technologies. In this study, we modified one top-performing molecular generation model on the basis of the generative pretraining transformer (GPT) architecture in three directions. Moreover, we propose an integrated end-to-end neural network learning framework based on one complete encoder-decoder architecture transformer model: Transfer Text-to-Text Transformer (T5), by learning the embedding vector representation space of conditional molecular properties to encode and guide the vector representation of SMILES sequences, resulting in the output of the final decoder block with a softmax output (maximum likelihood objective). Moreover, we evaluated the performance of these NLP-based generation models and another new model architecture based on a selective state space and selected the best approach jointing a transfer learning strategy for de novo drug discovery to target L858R/T790M/C797S-mutant EGFR in non-small cell lung cancer.

Digital Pathology and Ensemble Deep Learning for Kidney Cancer Diagnosis: Dartmouth Kidney Cancer Histology Dataset

Kidney cancer has become a major global health issue over time, showing how early detection can play a very important role in mediating the disease. Traditional histological image analysis is recognized as the clinical gold standard for diagnosis, although it is highly manual and labor-intensive. Due to this issue, many are interested in computer-aided diagnostic technologies to assist pathologists in their diagnostics. Specifically, deep learning (DL) has become a viable remedy in this field. Nonetheless, the capacity of existing DL models to extract comprehensive visual features for accurate classification is limited. Toward the end, this study proposes using ensemble models that combine the strengths of multiple transformers and deep learning model architectures. By leveraging the collective knowledge of these models, the ensemble enhances classification performance and enables more precise and effective kidney cancer detection. This study compares the performance of these suggested models to previous studies, all of which used the publicly accessible Dartmouth Kidney Cancer Histology Dataset. This study showed that the Vision Transformers, with an average accuracy of over 99%, were able to achieve high detection accuracy across all complete slide picture patches. In particular, the CAiT, DeiT, ViT, and Swin models outperformed ResNet. All things considered, the Vision Transformers consistently produced an average accuracy of 98.51% across all five-folds. These results demonstrated that Vision Transformers might perform well and successfully identify important features from smaller patches. Through utilizing histopathological images, our findings will assist pathologists in diagnosing kidney cancer, resulting in early detection and increased patient survival rates.

Real-time digital dermatitis detection in dairy cows on Android and iOS apps using computer vision techniques

Abstract The aim of the study was to deploy computer vision (CV) models for real-time detection of digital dermatitis (DD) lesions in cows using Android or iOS mobile applications. Early detection of DD lesions in dairy cows is crucial for prompt treatment and animal welfare. Android and iOS apps could facilitate routine and early DD detection in cows’ feet on dairy and beef farms. Upon detecting signs of DD, dairy farmers could implement preventive and treatment methods, including foot baths, topical treatment, hoof trimming, or quarantining cows affected by DD to prevent its spread. We applied transfer-learning to DD image data for 5 lesion classes, M0, M4H, M2, M2P, and M4P, on pretrained YOLOv5 model architecture using COCO-128 pretrained weights. The combination of localization loss, classification loss, and objectness loss was used for the optimization of prediction performance. The custom DD detection model was trained on 363 images of size 416 x 416 pixels and tested on 46 images. During model training, data were augmented to increase model robustness in different environments. The model was converted into TFLite format for Android devices and CoreML format for iOS devices. Techniques such as quantization were implemented to improve inference speed in real-world settings. The DD models achieved an average mAP (mean Average Precision) of 0.95 on the test dataset. When tested in real-time, iOS devices resulted in Cohen’s kappa value of 0.57 (95% CI: 0.49-0.65) averaged across the 5 lesion classes denoting the moderate agreement of the model detection with human investigators. The Android device resulted in a Cohen’s kappa value of 0.38 (95% CI: 0.29-0.47) denoting fair agreement between model and investigator. Combining M2 and M2P classes and M4H and M4P classes resulted in a Cohen’s kappa value of 0.65 (95% CI: 0.54-0.76) and 0.46 (95% CI: 0.35-0.57), for Android and iOS devices, respectively. For the two-class model (lesion vs. non-lesion), a Cohen’s kappa value of 0.74 (95% CI: 0.63-0.85) and 0.65 (95% CI: 0.52-0.78) was achieved for iOS and Android devices, respectively. iOS achieved a good inference time of 20 ms, compared to 57 ms on Android. Additionally, we deployed models on Ultralytics iOS and Android apps giving kappa scores of 0.56 (95% CI: 0.48-0.64) and 0.46 (95% CI: 0.37-0.55) respectively. Our custom iOS app surpassed the Ultralytics apps in terms of kappa score and the confidence score.

Magnetic skyrmion: from fundamental physics to pioneering applications

Skyrmionic devices exhibit energy-efficient and high-integration data storage and computing capabilities due to their small size, topological protection, and low drive current requirements. So, to realize these devices, an extensive study, from fundamental physics to practical applications, becomes essential. In this article, we present an exhaustive review of the advancements in understanding the fundamental physics behind magnetic skyrmions and the novel data storage and computing technologies based on them. We begin with an in-depth discussion of fundamental concepts such as topological protection, stability, statics, and dynamics essential for understanding skyrmions, henceforth the foundation of skyrmion technologies. For the realization of CMOS-compatible skyrmion functional devices, the writing and reading of the skyrmions are crucial. We discuss the developments in different writing schemes such as STT, SOT, and VCMA. The reading of skyrmions is predominantly achieved via two mechanisms: the magnetoresistive tunnel junction TMR effect and topological resistivity. So, a thorough investigation into the skyrmion Hall effect, topological properties, and emergent fields is also provided, concluding the discussion on skyrmion reading developments. Based on the writing and reading schemes, we discuss the applications of the skyrmions in conventional logic, unconventional logic, memory applications, and neuromorphic computing, including a model of a skyrmion-based SNN architecture achieving over 90% accuracy on MNIST and fashion-MNIST datasets. Furthermore, we outline the potential of skyrmion-hosting Majorana zero modes in emerging topological quantum computation and helicity-dependent skyrmion qubits.

A Survey of Research in Large Language Models for Electronic Design Automation

Within the rapidly evolving domain of Electronic Design Automation (EDA), Large Language Models (LLMs) have emerged as transformative technologies, offering unprecedented capabilities for optimizing and automating various aspects of electronic design. This survey provides a comprehensive exploration of LLM applications in EDA, focusing on advancements in model architectures, the implications of varying model sizes, and innovative customization techniques that enable tailored analytical insights. By examining the intersection of LLM capabilities and EDA requirements, the paper highlights the significant impact these models have on extracting nuanced understandings from complex datasets. Furthermore, it addresses the challenges and opportunities in integrating LLMs into EDA workflows, paving the way for future research and application in this dynamic field. Through this detailed analysis, the survey aims to offer valuable insights to professionals in the EDA industry, AI researchers, and anyone interested in the convergence of advanced AI technologies and electronic design.

Анализ изображений графиков зависимостей физических величин

Physical effects used at the conceptual design stage, described in “primary” sources of information, such as patents, often contain images of dependency graphs linking physical input and output quantities. Analysis of this information and its use to expand the description of a physical effect is a relevant task. The development of a method for analyzing graphic images for classifying dependency graphs of input and output physical quantities is described. This requires forming a labeled array of dependency graphs, as well as conducting computational experiments to identify the most effective architectures of neural network models. Algorithms for segmenting images of dependency graphs have been developed, allowing one to get rid of noisy (for the classification task) parts of the figure, such as coordinate axes, their designations, coordinate grids, etc.), the effectiveness of the OpenCV and scikit-image libraries has been tested on solving this problem. The formed labeled array contains more than 26 thousand images of dependency graphs. An algorithm for clustering images of dependency graphs by 9 classes (concave increase, concave decrease, convex increase, convex decrease, linear increase, linear decrease, constancy, jump increase, jump-like decrease) has been developed and implemented in software. Based on the results of the work, it can be concluded that all 3 methods of image clustering (LSTM, CNN and ViT) show almost the same results on the test dataset: Accuracy, Precision, Recall, F1-Score, AUC-ROC – 98%. At the same time, on arbitrary images from the patent array, the accuracy of the analysis decreases: for the LSTM and ViT methods by about 10%, and for CNN by about 2%.

Digital twin for the formal analysis of a depth of anesthesia controller

Effective management of depth of anesthesia (DoA) is crucial for patient safety in healthcare. Anesthesiologists typically adjust anesthetic dosages to maintain desired sedation, analgesia (pain relief), and muscle relaxation states. In this paper, we present a digital twin (DT) architecture for the formal modeling and verification of an infusion pump controller for DoA management. The DT incorporates a virtual patient model, an autonomous DoA controller adjusting the infusion rate of the anesthetic agent, i.e., propofol, a test-case manager, and a runtime verification monitor. Data exchange occurs via Ethernet frames. Challenges arise from noise in the Bispectral Index monitoring system readings and infusion rate measurements in clinical scenarios. To mitigate noise impact, we design a feedback controller that is robust against noise. We reason about DT performance by evaluating control specifications using a temporal-logic language within the context of our runtime verification tool.

Advanced Soil Moisture Predictive Methodology in the Maize Cultivation Region

Soil moisture is a critical determinant of the maize crop health and productivity. With over 60% of India's maize cultivation concentrated in South Indian states, accurately forecasting soil moisture is essential for optimizing irrigation and enhancing agricultural output. This study introduces an Improved Hybrid Machine Learning (IHML) model that integrates and optimizes Machine Learning (ML) models to deliver superior predictive performance. By leveraging data from key maize-growing districts in South India, the IHML model demonstrates enhanced convergence rates and accuracy compared to traditional ML approaches. The research framework is grounded in comprehensive correlation evaluations, which inform parameter selection and model architecture. Extensive comparisons reveal that the IHML model significantly outperforms individual ML models in forecasting soil moisture with higher precision. These findings highlight the potential of IHML models to advance smart farming practices and enable precise irrigation management, paving the way for improved crop yield and sustainable agriculture.

Enhancing sentiment analysis through deep layer integration with long short-term memory networks

This involves studying one of the most important parts of natural language processing (NLP): sentiment, or whether a thing that makes a sentence is neutral, positive, or negative. This paper presents an enhanced long short-term memory (LSTM) network for the sentiment analysis task using an additional deep layer to capture sublevel patterns from the word input. So, the process that we followed in our approach is that we cleaned the data, preprocessed it, built the model, trained the model, and finally tested it. The novelty here lies in the additional layer in the architecture of LSTM model, which improves the model performance. We added a deep layer with the intention of improving accuracy and generalizing the model. The results of the experiment are analyzed using recall, F1-score, and accuracy, which in turn show that the deep-layered LSTM model gives us a better prediction. The LSTM model outperformed the baseline in terms of accuracy, recall, and f1-score. The deep layer's forecast accuracy increased dramatically once it was trained to capture intricate sequences. However, the improved model overfitted, necessitating additional regularization and hyperparameter adjustment. In this paper, we have discussed the advantages and disadvantages of using deep layers in LSTM networks and their application to developing models for deep learning with better-performing sentiment analysis.

Investigating the performance of super-resolved remote sensing images on coastline segmentation with deep learning based methods

The use of satellite imagery in critical areas, such as environmental monitoring and natural disaster management, is becoming increasingly important. Applications like monitoring coastal areas, detecting coastal erosion, and tracking land use changes demand high accuracy and detailed analysis. Traditional methods for coastline segmentation are often limited by the low resolution (LR) and high complexity of satellite imagery. To address this challenge, Super Resolution (SR) algorithms are employed to enhance the resolution of satellite images, which is particularly beneficial when examining areas with intricate structures, such as coastlines. In this context, the integration of SR and segmentation techniques presents an innovative approach to achieving greater accuracy and efficiency in satellite image analysis. In this study, the resolution of satellite images was enhanced using the Super Resolution Generative Adversarial Networks (SRGAN) model. Thanks to the flexible architecture of the SRGAN model, it was successfully adapted to work with satellite images, yielding satisfactory results. Coastal segmentation was performed using low-resolution, super-resolved, and high-resolution Gokturk-1 (GT-1) satellite images, employing U-net, LinkNet, and DeepLabV3+ segmentation models for comparison. The results indicated that increment in image resolution significantly affects segmentation success. Additionally, better performance in coastline segmentation was achieved with U-net and LinkNet models. Although the DeepLabV3+ model is effective for segmentation, it tends to capture less detail compared to the other two models. Overall, the combination of SRGAN and the LinkNet segmentation model produced results that were closest to reality

Biomechanical Risk Classification in Repetitive Lifting Using Multi-Sensor Electromyography Data, Revised National Institute for Occupational Safety and Health Lifting Equation, and Deep Learning

Repetitive lifting tasks in occupational settings often result in shoulder injuries, impacting both health and productivity. Accurately assessing the biomechanical risk of these tasks remains a significant challenge in occupational ergonomics, particularly within manufacturing environments. Traditional assessment methods frequently rely on subjective reports and limited observations, which can introduce bias and yield incomplete evaluations. This study addresses these limitations by generating and utilizing a comprehensive dataset containing detailed time-series electromyography (EMG) data from 25 participants. Using high-precision wearable sensors, EMG data were collected from eight muscles as participants performed repetitive lifting tasks. For each task, the lifting index was calculated using the revised National Institute for Occupational Safety and Health (NIOSH) lifting equation (RNLE). Participants completed cycles of both low-risk and high-risk repetitive lifting tasks within a four-minute period, allowing for the assessment of muscle performance under realistic working conditions. This extensive dataset, comprising over 7 million data points sampled at approximately 1259 Hz, was leveraged to develop deep learning models to classify lifting risk. To provide actionable insights for practical occupational ergonomics and risk assessments, statistical features were extracted from the raw EMG data. Three deep learning models, Convolutional Neural Networks (CNNs), Multilayer Perceptron (MLP), and Long Short-Term Memory (LSTM), were employed to analyze the data and predict the occupational lifting risk level. The CNN model achieved the highest performance, with a precision of 98.92% and a recall of 98.57%, proving its effectiveness for real-time risk assessments. These findings underscore the importance of aligning model architectures with data characteristics to optimize risk management. By integrating wearable EMG sensors with deep learning models, this study enables precise, real-time, and dynamic risk assessments, significantly enhancing workplace safety protocols. This approach has the potential to improve safety planning and reduce the incidence and severity of work-related musculoskeletal disorders, ultimately promoting better health and safety outcomes across various occupational settings.

TransformerLSR: Attentive joint model of longitudinal data, survival, and recurrent events with concurrent latent structure.

In applications such as biomedical studies, epidemiology, and social sciences, recurrent events often co-occur with longitudinal measurements and a terminal event, such as death. Therefore, jointly modeling longitudinal measurements, recurrent events, and survival data while accounting for their dependencies is critical. While joint models for the three components exist in statistical literature, many of these approaches are limited by heavy parametric assumptions and scalability issues. Recently, incorporating deep learning techniques into joint modeling has shown promising results. However, current methods only address joint modeling of longitudinal measurements at regularly-spaced observation times and survival events, neglecting recurrent events. In this paper, we develop TransformerLSR, a flexible transformer-based deep modeling and inference framework to jointly model all three components simultaneously. TransformerLSR integrates deep temporal point processes into the joint modeling framework, treating recurrent and terminal events as two competing processes dependent on past longitudinal measurements and recurrent event times. Additionally, TransformerLSR introduces a novel trajectory representation and model architecture to potentially incorporate a priori knowledge of known latent structures among concurrent longitudinal variables. We demonstrate the effectiveness and necessity of TransformerLSR through simulation studies and analyzing a real-world medical dataset on patients after kidney transplantation.

AI And Deep Learning Models for Early Lung Cancer Detection Using Radiological Data

Lung cancer remains one of the deadliest cancers worldwide, with survival rates significantly dependent on early detection. Traditional diagnostic methods, while effective, often identify lung cancer in advanced stages, limiting treatment options. Recently, artificial intelligence (AI) and deep learning (DL) models have gained attention for their potential to assist in early lung cancer diagnosis through radiological data, particularly computed tomography (CT) and X-ray imaging. This review examines recent advancements in AI and DL models specifically applied to early lung cancer detection, offering a comprehensive look at model architecture, data preprocessing techniques, and performance metrics. AI models, particularly those using convolutional neural networks (CNNs) and recurrent neural networks (RNNs), have shown significant promise in improving diagnostic accuracy. In studies published over the past two years, CNN-based models have achieved sensitivity rates upwards of 90% when applied to large datasets, such as the LIDC-IDRI, a publicly available database containing thousands of annotated lung scans. Furthermore, hybrid approaches combining AI models with radiologist expertise have demonstrated reduced false positives, a frequent challenge in automated image analysis. A critical factor in these advancements has been data quality and volume. Large, annotated datasets with high-resolution images have enabled more robust training and validation, essential for refining these models. However, challenges remain, particularly regarding the standardization of data across institutions, variations in scanner quality, and ethical concerns surrounding patient privacy. The review also highlights studies exploring multi-modal approaches, integrating radiological data with clinical records and genetic markers to create more personalized diagnostic tools. These multi-modal methods have the potential to improve predictive accuracy further, though they currently require more extensive validation. Overall, while AI and DL technologies offer transformative potential for early lung cancer detection, widespread implementation depends on continued research in model accuracy, data standardization, and ethical safeguards. This paper concludes by emphasizing the need for collaboration across disciplines—AI researchers, radiologists, and oncologists—to refine these models into reliable tools that can be integrated seamlessly into clinical workflows for timely and accurate lung cancer diagnosis.

A deep learning model for assistive decision-making during robot-aided rehabilitation therapies based on therapists’ demonstrations

BackgroundA promising approach to improving motor recovery during rehabilitation is the use of robotic rehabilitation devices. These robotic devices provide tools to monitor the patient’s recovery progress while providing highly standardized and intensive therapy. A major challenge in using these robotic devices is the ability to decide when to assist the user. In this context, we propose a Deep Learning-based solution that can learn from a therapist’s criteria when a patient needs assistance during robot-aided rehabilitation therapy.MethodsAn experimental session was conducted with diverse people who suffered from neurological conditions. The participants used an upper limb rehabilitation robot to play a point-to-point game. A therapist supervised the robot-aided rehabilitation exercises and assisted the participants when considered necessary. This assistance provided by the therapist was detected to label those trajectories that were assisted to train a Deep Learning model that learns from the therapist when to assist. A series of transformations have been applied to the trajectories performed by the participants to generalize the method. Furthermore, the trajectory data was divided into sequences to be introduced to the model and continuously infer whether the user needs assistance. The data acquired during the experimental sessions have been divided into two datasets to train and evaluate the model: intra-participants (80% training, 20% validation) and test participants. The architecture of the Deep Learning model is conceived to perform time-series classification. It consists of diverse one-dimensional convolutional layers, a convolutional attention mechanism, and a Global Average Pooling layer. In addition, the output layer has one neuron with the sigmoid activation function, whose output can be interpreted as a probability of assistance. The model proposed in this study has been evaluated according to different metrics. In addition, the impact of applying fine-tuning to adapt the assistance to each user has been evaluated with the test participants.ResultsThe proposed model achieved an accuracy of 91.39% and an F1-Score of 75.15% with the validation dataset during a sequence-to-sequence evaluation, surpassing other state-of-the-art architectures. When evaluating the trajectories collected in the test dataset, the method proposed achieved an accuracy of 76.09% and an F1-Score of 74.42% after applying fine-tuning to each participant.ConclusionsThe results achieved by our Deep Learning-based method show the feasibility of learning assistance decision-making from experimented therapists. Furthermore, fine-tuning can be applied to personalize the assistance to each user and improve the accuracy of the method presented when deciding whether to assist with the rehabilitation robot.

Advances in diffusion models for image data augmentation: a review of methods, models, evaluation metrics and future research directions

Image data augmentation constitutes a critical methodology in modern computer vision tasks, since it can facilitate towards enhancing the diversity and quality of training datasets; thereby, improving the performance and robustness of machine learning models in downstream tasks. In parallel, augmentation approaches can also be used for editing/modifying a given image in a context- and semantics-aware way. Diffusion Models (DMs), which comprise one of the most recent and highly promising classes of methods in the field of generative Artificial Intelligence (AI), have emerged as a powerful tool for image data augmentation, capable of generating realistic and diverse images by learning the underlying data distribution. The current study realizes a systematic, comprehensive and in-depth review of DM-based approaches for image augmentation, covering a wide range of strategies, tasks and applications. In particular, a comprehensive analysis of the fundamental principles, model architectures and training strategies of DMs is initially performed. Subsequently, a taxonomy of the relevant image augmentation methods is introduced, focusing on techniques regarding semantic manipulation, personalization and adaptation, and application-specific augmentation tasks. Then, performance assessment methodologies and respective evaluation metrics are analyzed. Finally, current challenges and future research directions in the field are discussed.

Optimized convolutional neural network using African vulture optimization algorithm for the detection of exons

The detection of exons is an important area of research in genomic sequence analysis. Many signal-processing methods have been established successfully for detecting the exons based on their periodicity property. However, some improvement is still required to increase the identification accuracy of exons. So, an efficient computational model is needed. Therefore, for the first time, we are introducing an optimized convolutional neural network (optCNN) for classifying the exons and introns. The study aims to identify the best CNN model that provides improved accuracy for the classification of exons by utilizing the optimization algorithm. In this case, an African Vulture Optimization Algorithm (AVOA) is used for optimizing the layered architecture of the CNN model along with its hyperparameters. The CNN model generated with AVOA yielded a success rate of 97.95% for the GENSCAN training set and 95.39% for the HMR195 dataset. The proposed approach is compared with the state-of-the-art methods using AUC, F1-score, Recall, and Precision. The results reveal that the proposed model is reliable and denotes an inventive method due to the ability to automatically create the CNN model for the classification of exons and introns.