Deep Neural Network Research Articles

Exploring electron-beam induced modifications of materials with machine-learning assisted high temporal resolution electron microscopy

Directed atomic fabrication using an aberration-corrected scanning transmission electron microscope (STEM) opens new pathways for atomic engineering of functional materials. In this approach, the electron beam is used to actively alter the atomic structure through electron beam induced irradiation processes. One of the impediments that has limited widespread use thus far has been the ability to understand the fundamental mechanisms of atomic transformation pathways at high spatiotemporal resolution. Here, we develop a workflow for obtaining and analyzing high-speed spiral scan STEM data, up to 100 fps, to track the atomic fabrication process during nanopore milling in monolayer MoS2. An automated feedback-controlled electron beam positioning system combined with deep convolution neural network (DCNN) was used to decipher fast but low signal-to-noise datasets and classify time-resolved atom positions and nature of their evolving atomic defect configurations. Through this automated decoding, the initial atomic disordering and reordering processes leading to nanopore formation was able to be studied across various timescales. Using these experimental workflows a greater degree of speed and information can be extracted from small datasets without compromising spatial resolution. This approach can be adapted to other 2D materials systems to gain further insights into the defect formation necessary to inform future automated fabrication techniques utilizing the STEM electron beam.

An Artificial Intelligence-Assisted Flexible and Wearable Mechanoluminescent Strain Sensor System

HighlightsThe sandwich-structured flexible mechanoluminescent sensor (SFLC) film shows great application potential as wireless wearable strain sensor and encryption device.System-level integration of SFLC film with deep learning-based artificial intelligence enables fast and accurate interpretation of color data to strain values with automatic correction of errors caused by varying color temperatures.The smart glove wearable sensor based on the SFLC film combined with deep learning neural network enables fast and accurate hand gesture recognition.

A novel interpretative deep neural network with Grad-CAMs heatmap for the early diagnosis of Alzheimers disease

Abstract. In recent years, Alzheimers disease, a disease which targets the patients cognitive abilities and causes dementia, has become increasingly prevalent among the older population. Clinical practice today diagnoses the disease through MRI imaging and cognitive tests based on individual doctors personal experience. This reliance on varying doctoral experiences poses a huge challenge in during its early stages, when the symptoms are subtle and hard for clinicians to discern. As a result, many elderly today are only diagnosed when the disease has already progressed into a later, much obvious stage, significantly reducing their chance of receiving effective treatments. In our work, we applied a novel method in Alzheimers disease early diagnosis. Our unprecedented method uses a state-of-the-art image recognition model, a Vision Transformer (ViT), combined with a saliency map, Grad-CAM, to increase interpretability and alleviate the black box issue in the predicted results. We trained the ViT with ADNI imaging datasets and conducted similar experiments with other popular models for machine learning as baselines for performance comparison. The results show that our model has delivered the best performance out of all traditional methods, with an astonishing 99% accuracy. Compared to an average accuracy of 96% by other commonly used models for data prediction, our method has improved the final prediction accuracy to 3 standard deviation range, thereby accounting for almost all the possible outliers. These breakthroughs, paired with the website we created to increase the models reach, will prove to be a revolutionary contribution to the field of Alzheimers disease diagnosis.

Data discretization impact on deep learning for missing value imputation of continuous data

In various fields of information examination, for example, AI, profoundly getting the hang of missing information is a typical issue. Missing qualities should be tended to since they can adversely affect the exactness and adequacy of prescient models. This research investigates how data discretization affects deep learning methods for filling the missing values in datasets with continuous features. They provide a unique method for imputing missing values using deep neural networks (DNNs) called extravagant expectation maximization-deep neural network (EEM-DNN). This approach discretizes continuous features into separate intervals initially. This is justified by treating the issue of missing value imputation as a classification work, with the missing values being considered a distinct class. A DNN, designed explicitly for imputation, is then trained using the discretized data. The expectation maximization concepts are incorporated into the network architecture, and as a result, the network iteratively improves its imputation predictions. They run comprehensive experiments on several datasets from different fields to gauge the efficacy of the suggested strategy. The effectiveness of EEM-DNN is compared to that of other imputation approaches, such as traditional imputation techniques and deep learning methods without data discretization. Our findings show that data discretization significantly enhances imputation accuracy. In terms of imputation accuracy and prediction performance on downstream tasks, the EEM-DNN method regularly performs better than alternative methods. It also examines if various discretization techniques affect the overall imputation process. They find that the trade-off between bias and variance in imputed data depends on the discretization method selected. This highlights the significance of choosing a suitable discretization approach depending on the unique properties of the dataset.

Classification of Ship Type from Combination of HMM–DNN–CNN Models Based on Ship Trajectory Features

This study proposes an enhanced ship-type classification model that employs a sequential processing methodology integrating hidden Markov model (HMM), deep neural network (DNN), and convolutional neural network (CNN) techniques. Four different ship types—fishing boat, passenger, container, and other ship—were classified using multiple ship trajectory features extracted from the automatic identification system (AIS) and small fishing vessel tracking system. For model optimization, both ship datasets were transformed into various formats corresponding to multiple models, incorporating data enhancement and augmentation approaches. Speed over ground, course over ground, rate of turn, rate of turn in speed, berth distance, latitude/longitude, and heading were used as input parameters. The HMM–DNN–CNN combination was obtained as the optimal model (average F-1 score: 97.54%), achieving individual classification performances of 99.03%, 97.46%, and 95.83% for fishing boats, passenger ships, and container ships, respectively. The proposed approach outperformed previous approaches in prediction accuracy, with further improvements anticipated when implemented on a large-scale real-time data collection system.

Intraoperative Periprosthetic Fractures in Total Hip Arthroplasty: A 1.6-Million-Patient Analysis of Complications, Costs, and the Challenges in AI-Based Prediction.

Background: Periprosthetic fractures following total hip arthroplasty are serious complications occurring in up to 2.4% of primary cases, contributing to significant morbidity, extended hospital stays, and elevated healthcare costs. Predicting these fractures remains a challenge despite advances in surgical techniques and prosthetic materials. Methods: This study analyzed 1,634,615 cases of primary THA from the NIS database (2016-2019) using propensity score matching to compare outcomes between patients with and without intraoperative periprosthetic fractures. Predictive models, including logistic regression, decision tree, and deep neural network, were evaluated for their ability to predict fracture risk. Results: Patients with periprosthetic fractures exhibited a 14-fold increase in pulmonary embolism risk, a 12-fold increase in infections, and a 5-fold increase in hip dislocations. Fractures extended hospital stays (3.8 vs. 2.5 days) and added approximately USD 32,000 in costs per patient. The predictive models yielded low accuracy (AUC max = 0.605), underscoring the complexity of predicting periprosthetic fractures. Conclusions: Intraoperative periprosthetic fractures in THA significantly elevate complication rates, costs, and length of stay. Despite extensive modeling efforts, accurate prediction remains difficult, highlighting the need to focus on preventive strategies, such as improved surgical techniques and real-time intraoperative monitoring.

Understanding and modeling human-AI interaction of artificial intelligence tool in radiation oncology clinic using deep neural network: a feasibility study using three year prospective data

Objective.Artificial intelligence (AI) based treatment planning tools are being implemented in clinic. However, human interactions with such AI tools are rarely analyzed. This study aims to comprehend human planner's interaction with the AI planning tool and incorporate the analysis to improve the existing AI tool.Approach.An in-house AI tool for whole breast radiation therapy planning was deployed in our institution since 2019, among which 522 patients were included in this study. The AI tool automatically generates fluence maps of the tangential beams to create anAI plan. Human planner makes fluence edits deemed necessary and after attending physician approval for treatment, it is recorded asfinal plan. Manual modification value maps were collected, which is the difference between theAI-planand thefinal plan. Subsequently, a human-AI interaction (HAI) model using full scale connected U-Net was trained to learn such interactions and perform plan enhancements. The trained HAI model automatically modifies theAI planto generate AI-modified plans (AI-m plan), simulating human editing. Its performance was evaluated against originalAI-planandfinal plan. Main results. AI-m planshowed statistically significant improvement in hotspot control over theAI plan, with an average of 25.2cc volume reduction in breast V105% (p= 0.011) and 0.805% decrease in Dmax (p< .001). It also maintained the same planning target volume (PTV) coverage as thefinal plan, demonstrating the model has captured the clinic focus of improving PTV hot spots without degrading coverage.Significance.The proposed HAI model has demonstrated capability of further enhancing theAI planvia modeling human-AI tool interactions. This study shows analysis of human interaction with the AI planning tool is a significant step to improve the AI tool.

A dynamic neural network model for the identification of asbestos roofings in hyperspectral images covering a large regional area

AbstractAsbestos has been used extensively in several applications. Once it is known as a dangerous mineral, its usage has been prohibited and its identification and remediation play a very important role from the health safety point of view. Nowadays, deep learning techniques are used in many applications, especially for image analysis. They can be used to significantly reduce the time and cost of traditional detection methods. In this paper, taking advantage of asbestos spectral signature, a deep neural network is introduced in order to implement a complete methodology to identify asbestos roofings starting from hyperspectral images in a regional context. The novelty of the proposed approach is a dynamic mixing of models with different features, in order to accommodate classifications on widespread areas of both urban and rural territories. Indeed, the dataset used during the experiments described in this paper is a large one, consisting of many wide hyperspectral images with a geometric resolution of 1 m and with 186 bands, covering an entire region of approximately 8,000 . This is in contrast to other works in the literature where the analyzed areas are limited in size and uniform for physical features.

Learned scalable video coding for humans and machines

Video coding has traditionally been developed to support services such as video streaming, videoconferencing, digital TV, and so on. The main intent was to enable human viewing of the encoded content. However, with the advances in deep neural networks (DNNs), encoded video is increasingly being used for automatic video analytics performed by machines. In applications such as automatic traffic monitoring, analytics such as vehicle detection, tracking and counting, would run continuously, while human viewing could be required occasionally to review potential incidents. To support such applications, a new paradigm for video coding is needed that will facilitate efficient representation and compression of video for both machine and human use in a scalable manner. In this manuscript, we introduce an end-to-end learnable video codec that supports a machine vision task in its base layer, while its enhancement layer, together with the base layer, supports input reconstruction for human viewing. The proposed system is constructed based on the concept of conditional coding to achieve better compression gains. Comprehensive experimental evaluations conducted on four standard video datasets demonstrate that our framework outperforms both state-of-the-art learned and conventional video codecs in its base layer, while maintaining comparable performance on the human vision task in its enhancement layer.

DeepBBQ: A Deep Learning Approach to the Protein Backbone Reconstruction.

Coarse-grained models have provided researchers with greatly improved computational efficiency in modeling structures and dynamics of biomacromolecules, but, to be practically useful, they need fast and accurate conversion methods back to the all-atom representation. Reconstruction of atomic details may also be required in the case of some experimental methods, like electron microscopy, which may provide Cα-only structures. In this contribution, we present a new method for recovery of all backbone atom positions from just the Cα coordinates. Our approach, called deepBBQ, uses a deep convolutional neural network to predict a single internal coordinate per peptide plate, based on Cα trace geometric features, and then proceeds to recalculate the cartesian coordinates based on the assumption that the peptide plate atoms lie in the same plane. Extensive comparison with similar programs shows that our solution is accurate and cost-efficient. The deepBBQ program is available as part of the open-source bioinformatics toolkit Bioshell and is free for download and the documentation is available online.

Deep Learning for Urban Tree Canopy Coverage Analysis: A Comparison and Case Study

Urban tree canopy (UTC) coverage, or area, is an important metric for monitoring changes in UTC over large areas within a municipality. Several methods have been used to obtain these data, but remote sensing image classification is one of the fastest and most reliable over large areas. However, most studies have tested only one or two classification methods to accomplish this while using costly satellite imagery or LiDAR data. This study seeks to compare three urban tree canopy cover classifiers by testing a deep learning U-Net convolutional neural network (CNN), support vector machine learning classifier (SVM) and a random forests machine learning classifier (RF) on cost-free 2012 aerial imagery over a small southern USA city and midsize, growing southern USA city. The results of the experiment are then used to decide the best classifier and apply it to more recent aerial imagery to determine canopy changes over a 10-year period. The changes are subsequently compared visually and statistically with recent urban heat maps derived from thermal Landsat 9 satellite data to compare the means of temperatures within areas of UTC loss and no change. The U-Net CNN classifier proved to provide the best overall accuracy for both cities (89.8% and 91.4%), while also requiring the most training and classification time. When compared spatially with city heat maps, city periphery regions were most impacted by substantial changes in UTC area as cities grow and the outer regions get warmer. Furthermore, areas of UTC loss had higher temperatures than those areas with no canopy change. The broader impacts of this study reach the urban forestry managers at the local, state/province, and national levels as they seek to provide data-driven decisions for policy makers.

Biomechanical analysis and teaching strategies of complex movements in physical education teaching

This study suggests a new method for evaluating students’ ordinariness of movement in professional physical education by developing an assessment algorithm based on the biomechanical analysis of complex motions. The study aims to provide purposeful and data-driven techniques for assessing and optimizing movement ability in intricate physical tasks by utilizing higher motion capture and deep learning (DL) approaches, especially the Updated African Buffalo Optimization Based Deep Convolutional Neural Network (UABO-DCNN) categorization. The method includes collecting data utilizing high-precision movement capture equipment to research certain multifaceted movements, preprocessing trajectory data to extract kinematic, temporal, and spatial information, and increasing categorization algorithms with UABO-DCNN. The consequences specify that the algorithm can differentiate between normal and abnormal association patterns with outstanding accuracy. The UABO-DCNN model measures physical education teaching complex movements with accuracy (99.43%), precision (98.12%), recall (98.50%), F1-score (98.56%), and specificity (98.40%). Furthermore, the result is reliable, with a broader tendency toward instructive skill and individualized learning, which requires the development of physical education instruction actions by creating a culture of physical literacy and well-being. The implication of this employment includes an enhanced approach to promote optimal association skill increase in students, particularly for confronting complicated biomechanical measures.

Predicting Protein-Ligand Binding Affinity Using Fusion Model of Spatial-Temporal Graph Neural Network and 3D Structure-Based Complex Graph.

The investigation of molecular interactions between ligands and their target molecules is becoming more significant as protein structure data continues to develop. In this study, we introduce PLA-STGCNnet, a deep fusion spatial-temporal graph neural network designed to study protein-ligand interactions based on the 3D structural data of protein-ligand complexes. Unlike 1D protein sequences or 2D ligand graphs, the 3D graph representation offers a more precise portrayal of the complex interactions between proteins and ligands. Research studies have shown that our fusion model, PLA-STGCNnet, outperforms individual algorithms in accurately predicting binding affinity. The advantage of a fusion model is the ability to fully combine the advantages of multiple different models and improve overall performance by combining their features and outputs. Our fusion model shows satisfactory performance on different data sets, which proves its generalization ability and stability. The fusion-based model showed good performance in protein-ligand affinity prediction, and we successfully applied the model to drug screening. Our research underscores the promise of fusion spatial-temporal graph neural networks in addressing complex challenges in protein-ligand affinity prediction. The Python scripts for implementing various model components are accessible at https://github.com/ligaili01/PLA-STGCN.

Deep learning-based spike sorting: a survey.

Objective.Deep learning is increasingly permeating neuroscience, leading to a rise in signal-processing applications for extracellular recordings. These signals capture the activity of small neuronal populations, necessitating 'spike sorting' to assign action potentials (spikes) to their underlying neurons. With the rise in publications delving into new methodologies and techniques for deep learning-based spike sorting, it is crucial to synthesise these findings critically. This survey provides an in-depth evaluation of the approaches, methodologies and outcomes presented in recent articles, shedding light on the current state-of-the-art.Approach.Twenty-four articles published until December 2023 on deep learning-based spike sorting have been examined. The proposed methods are divided into three sub-problems of spike sorting: spike detection, feature extraction and classification. Moreover, integrated systems, i.e. models that detect spikes and extract features or do classification within a single network, are included.Main results.Although most algorithms have been developed for single-channel recordings, models utilising multi-channel data have already shown promising results, with efficient hardware implementations running quantised models on application-specific integrated circuits and field programmable gate arrays. Convolutional neural networks have been used extensively for spike detection and classification as the data can be processed spatiotemporally while maintaining low-parameter models and increasing generalisation and efficiency. Autoencoders have been mainly utilised for dimensionality reduction, enabling subsequent clustering with standard methods. Also, integrated systems have shown great potential in solving the spike sorting problem from end to end.Significance.This survey explores recent articles on deep learning-based spike sorting and highlights the capabilities of deep neural networks in overcoming associated challenges, but also highlights potential biases of certain models. Serving as a resource for both newcomers and seasoned researchers in the field, this work provides insights into the latest advancements and may inspire future model development.

Neural network-based predictions of antimicrobial resistance phenotypes in multidrug-resistant Acinetobacter baumannii from whole genome sequencing and gene expression.

Whole genome sequencing (WGS) potentially represents a rapid approach for antimicrobial resistance genotype-to-phenotype prediction. However, the challenge still exists to predict fully minimum inhibitory concentrations (MICs) and antimicrobial susceptibility phenotypes based on WGS data. This study aimed to establish an artificial intelligence-based computational approach in predicting antimicrobial susceptibilities of multidrug-resistant Acinetobacter baumannii from WGS and gene expression data. Antimicrobial susceptibility testing (AST) was performed using the broth microdilution method for 10 antimicrobial agents. In silico multilocus sequence typing (MLST), antimicrobial resistance genes, and phylogeny based on cgSNP and cgMLST strategies were analyzed. High-throughput qPCR was performed to measure the expression level of antimicrobial resistance (AMR) genes. Most isolates exhibited a high level of resistance to most of the tested antimicrobial agents, with the majority belonging to the IC2/CC92 lineage. Phylogenetic analysis revealed undetected transmission events or local outbreaks. The percentage agreements between AMR phenotype and genotype ranged from 70.08% to 89.96%, with the coefficient of agreement (κ) extending from 0.025 and 0.881. The prediction of AST employed by deep neural network models achieved an accuracy of up to 98.64% on the testing data set. Additionally, several linear regression models demonstrated high prediction accuracy, reaching up to 86.15% within an error range of one gradient, indicating a linear relationship between certain gene expressions and the corresponding antimicrobial MICs. In conclusion, neural network-based predictions could be used as a tool for the surveillance of antimicrobial resistance in multidrug-resistant A. baumannii.

Amp-Holo-Net enables speckle-free and high-quality coaxial amplitude-only holographic display

At present, amplitude-only holograms (AOH) are widely applied because of the high response speed and small pitch of amplitude spatial light modulators. However, AOHs are generally generated by traditional diffraction theories, and there are no effective deep learning algorithms to generate high-quality coaxial AOHs, which makes it difficult to achieve speckle-free, high-quality holographic displays using amplitude spatial light modulators. Thereby, we propose two cascaded deep neural networks, embedding physical models of the coaxial AOH to generate speckle-free and high-quality coaxial AOHs within an acceptable time, named Amp-Holo-Net, which is the first work that introduces deep learning to generate coaxial AOHs. We demonstrate the superiority of the proposed method through simulation, achieving an 81% reduction in speckle contrast of two-dimensional (2D) monochrome reconstruction and a 60% improvement in peak signal-to-noise ratio of 2D color reconstruction compared with point source method, and optical experiments are conducted to verify the validity of the proposed method. We believe our method fills the blank of AOH algorithms and paves the way for holographic displays based on amplitude spatial light modulators.

Adaptive hybrid prediction model for adapting to data distribution shifts in machining quality prediction

Abstract The accuracy of data-driven intelligent prediction for machining quality relies on the training samples. However, in actual applications, the continuous operation of machining equipment leads to gradual distribution shifts between the process data and the training samples for modeling. The shifts result in a degradation in the performance of predictive model, previous studies have often overlooked this issue. To tackle with the intricate problem, this research proposes a real-time model optimization approach. Firstly, a method for detecting machining data distribution shifts based on the two-sample Kolmogorov–Smirnov test is proposed. Then, an adaptive hybrid prediction model (AHPM) capable of real-time optimization is developed. This model consists of a deep neural network (DNN) and a broad learning system (BLS). DNN plays a primary role in prediction within the hybrid model with excellent generalization capability. BLS quickly completes optimization prior to DNN with its unique parameter update mechanism to compensate for prediction loss. Experimental results indicate that AHPM achieves the shortest optimization time while maintaining high accuracy, with post-optimization error reduction rates for mean squared error, mean absolute error, and mean absolute percentage error all exceeding 10%. In the test of application to actual machining cases, accuracy improved by 8.88% compared to traditional methods without optimization.

Task-agnostic exoskeleton control via biological joint moment estimation.

Lower-limb exoskeletons have the potential to transform the way we move1-14, but current state-of-the-art controllers cannot accommodate the rich set of possible human behaviours that range from cyclic and predictable to transitory and unstructured. We introduce a task-agnostic controller that assists the user on the basis of instantaneous estimates of lower-limb biological joint moments from a deep neural network. By estimating both hip and knee moments in-the-loop, our approach provided multi-joint, coordinated assistance through our autonomous, clothing-integrated exoskeleton. When deployed during 28 activities, spanning cyclic locomotion to unstructured tasks (for example, passive meandering and high-speed lateral cutting), the network accurately estimated hip and knee moments with an average R2 of 0.83 relative to ground truth. Further, our approach significantly outperformed a best-case task classifier-basedmethod constructed from splines and impedance parameters. When tested on ten activities (including level walking, running, lifting a 25 lb (roughly 11 kg) weight and lunging), our controller significantly reduced user energetics (metabolic cost or lower-limb biological joint work depending on the task) relative to the zero torque condition, ranging from 5.3 to 19.7%, without any manual controller modifications among activities. Thus, this task-agnostic controller can enable exoskeletons to aid users across a broad spectrum of human activities, a necessity for real-world viability.

Investigation on the mechanical uncertainty of resistance spot weldings and its effect on structure crashworthiness of automotive B-pillars under side impacts

Resistance spot welding (RSW) is the main connection method for the white body structure of automobiles. Understanding the mechanical uncertainty of RSWs and its effect on structure crashworthiness design is of great significance for vehicle safety and lightweight design. In this work, the mechanical uncertainty of RSW was experimentally investigated and its effect on automotive B-pillar crashworthiness under side impacts was numerically analyzed based on a hybrid response surface model (HRSM). Firstly, the mechanical uncertainty of RSWs and associated failure mechanism were investigated and the fluctuation range of RSW strength under different loading conditions were quantified according to experimental results. Then, a reasonable RSW modeling method and subsystem collision simulation approach were demonstrated and the univariate sensitivity analysis was carried out to reveal the effect of RSWs’ mechanical property on B-pillar crashworthiness. Finally, a HRSM integrating Kriging polynomial and deep neural network (DNN) were proposed as surrogate model, based on which the multi-factor analysis was conducted to explore effects of RSW strength uncertainty on structure crashworthiness design.

Advanced Computational Intelligence Techniques for Real-Time Decision-Making in Autonomous Systems

This research explores advanced computational intelligence techniques aimed at enhancing real-time decision-making in autonomous systems. The increasing reliance on autonomous technologies across sectors such as transportation, healthcare, and industrial automation demands robust, adaptive, and reliable decision-making frameworks. This study introduces a novel hybrid model that integrates Reinforcement Learning (RL), Deep Neural Networks (DNN), and Fuzzy Logic to enable autonomous systems to make accurate and timely decisions in complex, dynamic environments. The proposed framework leverages RL for adaptive decision-making, DNNs for pattern recognition and prediction, and Fuzzy Logic for handling uncertainty in system states. Experimental evaluations were conducted using high-fidelity simulations across three scenarios: autonomous vehicle navigation, real-time patient monitoring in healthcare, and robotic process automation. Results indicate a 25% improvement in decision accuracy, a 30% reduction in response time, and enhanced robustness against environmental variability compared to conventional decision-making methods. The findings underscore the effectiveness of computational intelligence in supporting critical decisions in real-time, marking a significant step toward more capable and reliable autonomous systems.