Domain Alignment Research Articles

Gradient Harmonization in Unsupervised Domain Adaptation.

Unsupervised domain adaptation (UDA) intends to transfer knowledge from a labeled source domain to an unlabeled target domain. Many current methods focus on learning feature representations that are both discriminative for classification and invariant across domains by simultaneously optimizing domain alignment and classification tasks. However, these methods often overlook a crucial challenge: the inherent conflict between these two tasks during gradient-based optimization. In this paper, we delve into this issue and introduce two effective solutions known as Gradient Harmonization, including GH and GH++, to mitigate the conflict between domain alignment and classification tasks. GH operates by altering the gradient angle between different tasks from an obtuse angle to an acute angle, thus resolving the conflict and trade-offing the two tasks in a coordinated manner. Yet, this would cause both tasks to deviate from their original optimization directions. We thus further propose an improved version, GH++, which adjusts the gradient angle between tasks from an obtuse angle to a vertical angle. This not only eliminates the conflict but also minimizes deviation from the original gradient directions. Finally, for optimization convenience and efficiency, we evolve the gradient harmonization strategies into a dynamically weighted loss function using an integral operator on the harmonized gradient. Notably, GH/GH++ are orthogonal to UDA and can be seamlessly integrated into most existing UDA models. Theoretical insights and experimental analyses demonstrate that the proposed approaches not only enhance popular UDA baselines but also improve recent state-of-the-art models.

GALA: Graph Diffusion-based Alignment with Jigsaw for Source-free Domain Adaptation.

Source-free domain adaptation is a crucial machine learning topic, as it contains numerous applications in the real world, particularly with respect to data privacy. Existing approaches predominantly focus on Euclidean data, such as images and videos, while the exploration of non-Euclidean graph data remains scarce. Recent graph neural network (GNN) approaches could suffer from serious performance decline due to domain shift and label scarcity in source-free adaptation scenarios. In this study, we propose a novel method named Graph Diffusion-based Alignment with Jigsaw (GALA) tailored for source-free graph domain adaptation. To achieve domain alignment, GALA employs a graph diffusion model to reconstruct source-style graphs from target data. Specifically, a score-based graph diffusion model is trained using source graphs to learn the generative source styles. Then, we introduce perturbations to target graphs via a stochastic differential equation instead of sampling from a prior, followed by the reverse process to reconstruct source-style graphs. We feed them into an off-the-shelf GNN and introduce class-specific thresholds with curriculum learning, which can generate accurate and unbiased pseudo-labels for target graphs. Moreover, we develop a simple yet effective graph mixing strategy named graph jigsaw to combine confident graphs and unconfident graphs, which can enhance generalization capabilities and robustness via consistency learning. Extensive experiments on benchmark datasets validate the effectiveness of GALA. The source code is available at https://github.com/luo-junyu/GALA.

Investigation of the effect of crystal size on the piezoelectric features of lead-free barium titanate ceramic using molecular dynamics simulation.

This study investigated the piezoelectric properties of BaTiO3 ceramics with different sizes through molecular dynamics simulations. The results show that all samples reached thermal equilibrium at 300K and equilibrium in potential energy within 10ns, confirming effective equilibration. As the size of the ceramics increased, the mean square displacement and diffusion coefficients decreased from 0.217 and 0.0034 to 0.1934 Å2 and 0.003 Å2/ns, attributed to a more uniform microstructure with fewer defects, resulting in reduced ion mobility. Furthermore, saturation polarization, residual polarization, and coercive field values increased from 0.35, 0.1, and 0.175 to 0.42 C/m2, 0.16 C/m2, and 0.282 MV/m, respectively, with increasing sample size, highlighting enhanced polarization responses due to a greater volume of ferroelectric material. Larger barium titanate (BaTiO3) crystals can have better polarization due to more domains aligning, but they may not deform as much (lower strain) because the walls among those domains can't move freely. While improved domain alignment contributed to higher polarization, the increased stress can restrict the mobility of the domain walls. These findings provided valuable insights into the size-dependent behavior of BaTiO3 ceramics, essential for optimizing their applications in electronic devices and sensors. The study underscored the importance of understanding microstructural effects on material properties for future advancements in ferroelectric technology.

A class alignment network based on self-attention for cross-subject EEG classification.

Due to the inherent variability in EEG signals across different individuals, domain adaptation and adversarial learning strategies are being progressively utilized to develop subject-specific classification models by leveraging data from other subjects. These approaches primarily focus on domain alignment and tend to overlook the critical task-specific class boundaries. This oversight can result in weak correlation between the extracted features and categories. To address these challenges, we propose a novel model that uses the known information from multiple subjects to bolster EEG classification for an individual subject through adversarial learning strategies. Our method begins by extracting both shallow and attention-driven deep features from EEG signals. Subsequently, we employ a class discriminator to encourage the same-class features from different domains to converge while ensuring that the different-class features diverge. This is achieved using our proposed discrimination loss function, which is designed to minimize the feature distance for samples of the same class across different domains while maximizing it for those from different classes. Additionally, our model incorporates two parallel classifiers that are harmonious yet distinct and jointly contribute to decision-making. Extensive testing on two publicly available EEG datasets has validated our model's efficacy and superiority.

Efficient Cross-Domain Fault Diagnosis via Distributed Multi-Source Domain Deep Transfer Learning

Abstract In the actual industrial production, the working conditions of rotating machinery are complex and changeable, and the health-state monitoring data are increasingly large and difficult to be labeled, which will seriously restrict the accuracy and efficiency of cross-domain fault diagnosis (CDFD) of rotating machinery. Therefore, an efficient multi-source domain deep transfer learning (MDDTL) method for CDFD of rotating machinery is proposed. Firstly, a MDDTL model is constructed to improve the accuracy of CDFD. In the model, a dual-phase domain alignment strategy is designed, which considers the alignment of feature distributions between each source and target domain pair in the feature space and that of prediction probabilities between domain-specific fault classifiers in the output space. The fault prediction results from multiple different fault classifiers are merged dynamically by the proposed imbalanced adaptive prediction strategy. Secondly, a data-parallel distributed training scheme for MDDTL model is proposed. Based on the idea of data parallelism, the distributed parallel training of MDDTL model is performed with Horovod-GPU platform, and the parameters are synchronously updated with the bandwidth-optimal Ring-AllReduce architecture. Under the premise of ensuring the accuracy of fault diagnosis (FD), the training time of MDDTL model is significantly reduced. Finally, extensive experiments are conducted to verify the effectiveness of the proposed MDDTL method. The results demonstrate that the proposed method not only effectively improves the accuracy of CDFD of rotating machinery, but also significantly improves the training efficiency of MDDTL model. After adopting the proposed method, the diagnosis accuracies achieved under two different cross-working condition scenarios reach 97.09% and 97.87% respectively, and the model training time is reduced by 73.62% when facing a large-scale rotating machinery training set.

A Three-Stage Cross Domain Intelligent Fault Diagnosis Method for Multiple New Faults

Abstract Due to the randomness and complexity of mechanical faults, the new fault modes usually occur unexpectedly in the actual scenarios. In response to the challenges, a three-stage crossing domain intelligent fault diagnosis method is presented in article. Firstly, the partial domain alignment is achieved based on improved target weighted mechanism, and the outlier identifier is constructed to automatically separate the new fault classes. Then, the unsupervised learning model with silhouette coefficient are built to determine the number of new fault categories. Lastly, the simulation signals are further adopted to determine the specific fault categories. Sufficient experiments on the axial piston pump and public bearing datasets validate that the proposed method could predict the number of new fault categories and judge the specific fault categories. The results indicate that the proposed method outperforms the other methods and has promising practical application in fault diagnosis with multiple new faults.

Contrastive learning-enabled digital twin framework for fault diagnosis of rolling bearing

Abstract Rolling bearings are essential components in various industrial machines, and their failures can lead to significant downtime and maintenance costs. Traditional data-driven fault diagnosis methods often require extensive fault datasets for training, which may not always be available in critical industrial scenarios, limiting their practicality. Digital twins, virtual representations of physical entities reflecting their operational conditions, offer a promising solution for the fault diagnosis of rolling bearings with limited fault data. In this paper, we propose a novel digital twin-driven framework to address the challenge of limited training data in rolling bearing fault diagnosis. Firstly, a virtual bearing simulation model is used to generate the simulated data. Subsequently, a transformer-based network is introduced to learn the discrepancy features from the raw data. Then, a Maximum Mean Discrepancy loss and a supervised contrastive learning loss for raw and augmentation data are established to achieve global domain alignment and instance-based domain alignment. Finally, an unsupervised contrastive learning loss for the augmentation data of the target domain is established to further improve the diagnostic performance. In five cross-domain fault diagnosis tasks representing real industrial scenarios set in this study, the proposed method achieved an average diagnostic accuracy of 84.40%, exceeding two existing advanced domain adaptation methods by more than 10%. The experimental results demonstrate that the proposed method achieves high diagnostic performance in real industrial scenarios where labeled data is lacking. This shows its significant benefits for monitoring the condition of critical bearings.

Cry1Ac toxin binding in the velvetbean caterpillar Anticarsia gemmatalis: study of midgut aminopeptidases N.

The velvetbean caterpillar Anticarsia gemmatalis is one of the main soybean defoliators in Brazil. Currently, the main biopesticide used to control insect pests worldwide is the bacteria Bacillus thuringiensis (Bt), which produces entomopathogenic Crystal toxins (Cry) that act in the midgut of susceptible insects, leading them to death. The mode of action of Cry toxins in the midgut involves binding to specific receptors present on the brush border of epithelial cells such as aminopeptidase N (APN), alkaline phosphatase (ALP), cadherin, and others. Mutations in these receptors, among other factors, may be involved in the development of resistance; identification of functional Cry receptors in the midgut of A. gemmatalis is crucial to develop effective strategies to overcome this possible scenario. This study's goal is to characterize APNs of A. gemmatalis and identify a receptor for Cry1Ac in the midgut. The interaction of Bt spores with the midgut epithelium was observed in situ by immunohistochemistry and total aminopeptidase activity was estimated in brush border membrane vesicle (BBMV) samples, presenting higher activity in challenged individuals than in control ones. Ten APN sequences were found in a A. gemmatalis' transcriptome and subjected to different in silico analysis, such as phylogenetic tree, multiple sequence alignment and identification of signal peptide, activity domains and GPI-anchor signal. BBMV proteins from 5th instar larvae were submitted to a ligand blotting using activated Cry1Ac toxin and a commercial anti-Cry polyclonal antibody; corresponding bands of proteins that showed binding to Cry toxin were excised from the SDS-PAGE gel and subjected to mass spectrometry analysis, which resulted in the identification of seven of those APNs. Quantitative PCR was realized to compare expression levels between individuals subjected to sublethal infection with Bt spores and control ones, presenting up- and downregulations upon Bt infection. From these results, we can infer that aminopeptidases N in A. gemmatalis could be involved in the mode of action of Cry toxins in its larval stage.

Essentials of genomics in nursing undergraduate education: A discussion paper

AimTo map the 2021 American Association of Colleges of Nursing Essentials to the American Nurses Association Essentials of Genomic Nursing for all nurses and provide resources for nursing faculty to support the seamless integration of genomics into existing undergraduate curricula. BackgroundSince the completion of the Human Genome Project in 2003, rapid advancements in genomic science leading to practical applications of genomics have revolutionized all areas of healthcare. Nursing is built on foundational life sciences, including genomics. As the largest segment of the healthcare workforce, who spend the most time with patients and families, nurses play a critical role in healthcare teams integrating genomic knowledge into patient care to improve health and well-being. Consequently, nurses must be equipped with foundational genomic knowledge and skills during their undergraduate education. However, there is wide variability in whether and how nursing programs have incorporated genomics into their curricula. Additionally, nursing faculty may have limited knowledge of foundational genomic concepts and lack confidence in teaching genomics. DesignDiscussion paper MethodsWe aligned domains from the American Association of Colleges of Nursing Essentials and American Nurses Association Essentials of Genomic Nursing. ResultsA map illustrating alignment in multiple areas, which provide examples of ways to integrate genomics into existing nursing curricula. ConclusionAlthough based on domains developed in the United States, the map, curricular resources, example learning outcomes, and clinical vignettes can be used by nursing faculty globally to prepare future nurses who are competent in providing genomics-informed nursing care on entry-to-practice.

Source free domain adaptation for kidney and tumor image segmentation with wavelet style mining

Kidney cancer is a serious malignant disease, and early diagnosis along with precise segmentation are crucial for effective treatment. However, due to the scarcity of labelled medical image data, the development of the intelligent diagnosis of kidney cancer is restricted. To address this challenge, we propose a novel unsupervised domain adaptation (UDA) framework specifically designed for kidney and tumor CT image segmentation. Our framework consists of a generation phase and an adaptation phase. In the generation phase, we employ a wavelet-based style mining generator to create class-specific source-like images, facilitating domain alignment. In the adaptation phase, we introduce contrastive domain extraction and compact-aware domain consistency modules, enhancing feature-level and output-level adaptability through data augmentation techniques. Experimental results demonstrate that our method performs better in kidney and tumor segmentation tasks, exhibiting higher accuracy and generalization capability than state-of-the-art domain adaptation methods. This indicates that our approach has significant advantages in medical image segmentation for kidneys and tumors.

The Framework of GeoSOT-3D Grid Modeling for Spatial Artificial Intelligence

Abstract. With the development of society, spatial artificial intelligence (spatial AI) research is gradually able to play a greater role. However, spatial AI has problems such as data alignment, poor interpretability, and cross domain learning. Therefore, this paper proposes an innovative GeoSOT-3D grid modeling framework for spatial AI research, which enhances the application capabilities of spatial AI. Grid modeling will be able to run through the upstream and downstream of spatial AI research, providing encoding calculations and spatial neighborhood embedding matrices for spatial data. This paper also uses task examples to demonstrate how to effectively organize and index spatial data using GeoSOT-3D grids and conduct spatial AI research. The use of GeoSOT-3D grids for spatial AI analysis has enormous potential and broad application prospects, which will help promote the further development and application of spatial AI.

ADA-UDA: A transferable transformer framework for rumor detection using Adversarial Domain Alignment within Unsupervised Domain Adaptation

The rapid spread of misinformation on social media has posed significant challenges, particularly in the early detection of rumors, which is critical to mitigating their negative impact. However, the limited data available during the initial stages of trending topics poses significant challenges, including analysis constraints and model overfitting. Therefore, the adoption of specific methods are needed to identify patterns from historical sample data and enable effective knowledge transfer, facilitating inductive reasoning with limited data on new topics. To address these challenges, we propose an Adversarial Domain Alignment Unsupervised Domain Adaptation (ADA-UDA) method based on the Transformer architecture. Our approach leverages labeled historical data from the source domain alongside a limited quantity of unlabeled data from new trending topics in the target domain. At the heart of our method is the Parameter Transferable Module (PTM), which guides the Transformer to focus on transferable and distinguishable features, thereby enhancing the model’s ability to perform effective inductive reasoning with limited data. We conducted extensive experiments to evaluate our method, benchmarking it against current mainstream rumor detection techniques. The findings indicate that our ADA-UDA method outperforms existing approaches, underscoring its potential for early and accurate rumor detection in emerging topics.

USMDA: Unsupervised Multisource Domain Adaptive ADHD prediction model using neuroimaging

There is an increasing number of large-scale cross-site database collections of neuroimaging markers (sMRI and fMRI) for studying neurodevelopmental illnesses (NDDs). Although a huge amount of data favors machine learning-based categorization algorithms, the unique heterogeneity of each site can impair cross-site generalization capacity. It is critical to create Unsupervised domain adaption methods for NDDs because obtaining appropriate diagnoses or labeling for NDDs might be problematic. In our work, we focus on Attention-deficit/hyperactivity disorder, which is the most common and frequently co-occurring NDD. We present an unsupervised multisource domain adaptation network (USMDA) with four primary components: Domain Alignment Module, Discrepancy Estimator, Pre-trained Model Generator, and Unsupervised Network. The Domain Alignment module is intended to incrementally and effectively align graph representations of the source and target domains. At the same time, the binary cross entropy regularizer is introduced for the first time during the training of a model learned on multiple source domains to improve existing feature alignment methods such as Transfer Joint Matching (TJM) and Joint Distribution Adaptation (JDA) by learning good unsupervised features. In an unsupervised network, the grid search optimization technique generates the optimal pseudo labels for unlabeled target data. We validate our proposed technique first on existing feature-level DA methods such as JDA, TJM, and Correlation alignment (CORAL), on the publically accessible dataset ADHD-200 and then by using binary cross-entropy in existing DA methods such as TJMCE and JDACE. The experimental results show that our proposed USMDAJDACE method when applied to multisite sMRI and fMRI ADHD data, can significantly outperform competitive methods for multi-center ADHD diagnosis.

DCST: Dual Cross-Supervision for Transformer-based Unsupervised Domain Adaptation

Unsupervised Domain Adaptation aims to leverage a source domain with ample labeled data to tackle tasks on an unlabeled target domain. However, this poses a significant challenge, particularly in scenarios exhibiting significant disparities between the two domains. Prior methods often fall short in challenging domains due to the impact of incorrect pseudo-labeling noise and the limits of handcrafted domain alignment rules. In this paper, we propose a novel method called DCST (Dual Cross-Supervision Transformer), which improves upon existing methods in two key aspects. Firstly, vision transformer is combined with dual cross-supervision learning strategy to enforce consistency learning from different domains. The network accomplishes domain-specific self-training and cross-domain feature alignment in an adaptive manner. Secondly, due to the presence of noise in challenging domain, and the need to reduce the risks of model collapse and overfitting, we propose a Domain Shift Filter. Specifically, this module allows the model to leverage the memory of source domain features to facilitate a smooth transition. It can also improve the effectiveness of knowledge transfer between domains with significant gaps. We conduct extensive experiments on four benchmark datasets and achieved the best classification results, including 94.3% on Office-31, 86.0% on Office-Home, 89.3% on VisDA-2017, and 48.8% on DomainNet. Code is available in https://github.com/Yislight/DCST.

Optimized magnetic properties of a Terfenol-D alloy in an extended temperature range using a high magnetic field

Tb–Dy–Fe alloys are among the most suitable magnetostrictive materials for high-power transducers. Optimizing magnetic properties in an extended temperature range could ensure the stable operation of transducers. In this work, a high magnetic field is applied to the directional solidification of Tb–Dy–Fe alloys. We study the microstructure, crystallographic orientation, magnetic susceptibility, crystal structure, and magnetic domain of samples. When the content and alignment of the magnetic phase along with crystallographic orientation remain basically invariant, the magnetic susceptibility of samples increases with the magnetic flux density of the high magnetic field throughout the temperature range from 273 K to Curie temperature (T C). At 4 T, the maximum magnetic susceptibility is increased by ∼ 40% compared with the sample without a high magnetic field applied, and the advantage is maintained in the range ∼ 300 K. Analysis shows that the enhancement of magnetic susceptibility is not due to the change in crystal structure, as commonly believed, but to the highly ordered alignment of magnetic domains. This research provides a new method for improving the temperature properties of magnetic materials using a high magnetic field.

Multifunctional Nano‐Conductive Hydrogels With High Mechanical Strength, Toughness and Fatigue Resistance as Self‐Powered Wearable Sensors and Deep Learning‐Assisted Recognition System

AbstractHigh mechanical strength, toughness, and fatigue resistance are essential to improve the reliability of conductive hydrogels for self‐powered sensing. However, achieving mutually exclusive properties simultaneously remains challenging. Hence, a novel directed interlocking strategy based on topological network structure and mechanical training is proposed to construct tough hydrogels by optimizing the network structure and modulating the orientation of molecular chains. Combining Zn2+ crosslinked cellulose nanofibers (CNFs) and a polyacrylamide‐poly(vinyl alcohol) double‐network, the unique interlocked‐network structure exhibits an enhanced toughening effect due to hydrogen bonding and metal‐ligand interactions. The aligned nanocrystalline domains achieved by training further contribute to an increase in the toughness and fatigue thresholds. This innovative approach synergistically enhances the mechanical properties of the nano‐conductive hydrogel, achieving a maximum tensile strength of 4.98 MPa and a toughness of 48 MJ m−3. Notably, the CNFs template with anchored polyaniline, when oriented through mechanical training, forms a unique directional conductive pathway, which significantly enhances the power output performance. Besides, a motion recognition system based on a self‐powered sensing device is designed with the assistance of deep learning techniques to accurately identify human motion behaviors. This work showcases a potentially transformative flexible electronic material for self‐powered sensing systems and intelligent recognition systems.

Towards Discriminative Class-Aware Domain Alignment via Coding Rate Reduction for Unsupervised Adversarial Domain Adaptation

Unsupervised domain adaptation (UDA) methods, based on adversarial learning, employ the means of implicit global and class-aware domain alignment to learn the symmetry between source and target domains and facilitate the transfer of knowledge from a labeled source domain to an unlabeled target domain. However, these methods still face misalignment and poor target generalization due to small inter-class domain discrepancy and large intra-class discrepancy of target features. To tackle these challenges, we introduce a novel adversarial learning-based UDA framework named Coding Rate Reduction Adversarial Domain Adaptation (CR2ADA) to better learn the symmetry between source and target domains. Integrating conditional domain adversarial networks with domain-specific batch normalization, CR2ADA learns robust domain-invariant features to implement global domain alignment. For discriminative class-aware domain alignment, we propose the global and local coding rate reduction methods in CR2ADA to maximize inter-class domain discrepancy and minimize intra-class discrepancy of target features. Additionally, CR2ADA combines minimum class confusion and mutual information to further regularize the diversity and discriminability of the learned features. The effectiveness of CR2ADA is demonstrated through experiments on four UDA datasets. The code can be obtained through email or GitHub.

Multi-source domain adaptation network for partial discharge severity assessment in gas-insulated switchgear

Abstract Gas-insulated switchgear (GIS) partial discharge (PD) severity assessment is critical for ensuring the reliable operation of GIS systems. However, existing assessment methods often overlook the long-term dependencies of historical data and fail to adequately address challenges related to limited on-site samples and variations in sample distribution. To overcome these challenges, we propose a novel multi-source domain adaptation network (MSDAN) specifically designed for on-site GIS PD severity assessment, which is the first model developed considering distribution differences in GIS PD severity assessment for different defect types. Our approach begins with the development of a feature extractor that captures both discernible PD features and long-term dependencies. We then introduce a multi-source domain adaptation strategy to mitigate distribution disparities across PD severity samples from different defect types, ensuring effective domain alignment. Additionally, we incorporate an adaptive weighted classification mechanism to accurately assess PD severity by considering the varying contributions of different defect types to the target domain task. Experimental results demonstrate that MSDAN achieves a remarkable accuracy of 95.38% in on-site GIS PD severity assessment, outperforming other benchmark models in both accuracy and robustness. This highlights the potential of MSDAN as a robust solution for real-world GIS PD severity assessment.

TriLA: Triple-Level Alignment Based Unsupervised Domain Adaptation for Joint Segmentation of Optic Disc and Optic Cup.

Cross-domain joint segmentation of optic disc and optic cup on fundus images is essential, yet challenging, for effective glaucoma screening. Although many unsupervised domain adaptation (UDA) methods have been proposed, these methods can hardly achieve complete domain alignment, leading to suboptimal performance. In this paper, we propose a triple-level alignment (TriLA) model to address this issue by aligning the source and target domains at the input level, feature level, and output level simultaneously. At the input level, a learnable Fourier domain adaptation (LFDA) module is developed to learn the cut-off frequency adaptively for frequency-domain translation. At the feature level, we disentangle the style and content features and align them in the corresponding feature spaces using consistency constraints. At the output level, we design a segmentation consistency constraint to emphasize the segmentation consistency across domains. The proposed model is trained on the RIGA+ dataset and widely evaluated on six different UDA scenarios. Our comprehensive results not only demonstrate that the proposed TriLA substantially outperforms other state-of-the-art UDA methods in joint segmentation of optic disc and optic cup, but also suggest the effectiveness of the triple-level alignment strategy.

Integrating multimodal contrastive learning with prototypical domain alignment for unsupervised domain adaptation of time series

Unsupervised domain adaptation (UDA) addresses the challenge of transferring knowledge from a labeled source domain to an unlabeled target domain. This task is particularly critical for time series data, characterized by unique temporal dynamics. However, existing methods often fail to capture these temporal dependencies, leading to domain discrepancies and loss of semantic information. In this study, we propose a novel framework for the unsupervised domain adaptation of time series (UDATS) that integrates Multimodal Contrastive Adaptation (MCA) and Prototypical Domain Alignment (PDA). MCA leverages image encoding techniques and prompt learning to capture complex temporal patterns while preserving semantic information. PDA constructs multimodal prototypes, combining visual and textual features to align target domain samples accurately. Our framework demonstrates superior performance across various application domains, including human activity recognition, mortality prediction, and fault detection. Experiments show our method effectively addresses domain discrepancies while preserving essential semantic content, outperforming state-of-the-art models.