Feature Embedding Research Articles

Study on the tribological performance of self-lubricating thrust ball bearings with different embedded features

PurposeThe purpose of this paper is to develop a new type of embedded solid self-lubricating thrust ball bearing for conditions where grease lubrication cannot be used and to analyze its tribological performance under different lubrication characteristics (lubrication position, width and filling amount).Design/methodology/approachLubrication parameters such as position (a), width (W) and filling amount (Q) were considered. Grooves were made on the raceway with a fiber laser and solid self-lubricating materials were applied through scraping. The frictional behavior of the new bearing was analyzed using a vertical test rig and the bearing’s surface topography was examined with a noncontact profilometer to study wear mechanisms.FindingsThe new inlay thrust ball bearings exhibited excellent lubrication effects and effectively controlled the temperature rise of the bearings. When a is 0 degrees, W is 0.5 mm and Q is 16 mg, the bearing experiences the least wear, and the friction coefficient and temperature are the lowest, measuring 0.001 and 41.52 degrees, respectively. Under the same experimental conditions, compared to smooth bearings without solid lubrication, the friction coefficient decreased by 96.88% and the temperature decreased by 59.74%.Originality/valueThis study presents a self-lubricating thrust ball bearing designed for conditions where grease lubrication is not feasible. A comprehensive investigation was conducted on its surface morphology, wear mechanisms and tribological performance. This work provides valuable insights into the research of self-lubricating thrust ball bearings.Peer reviewThe peer review history for this article is available at: https://publons.com/publon/10.1108/ILT-03-2024-0073/

Few-shot online anomaly detection and segmentation

Detecting anomaly patterns from images is a crucial artificial intelligence technique in industrial applications. Recent research in this domain has emphasized the necessity of a large volume of training data, overlooking the practical scenario where, post-deployment of the model, unlabeled data containing both normal and abnormal samples can be utilized to enhance the performance. Consequently, this paper focuses on addressing the challenging yet practical few-shot online anomaly detection and segmentation (FOADS) task. Under the FOADS framework, models are trained on a few-shot normal dataset, followed by inspection and improvement of their capabilities by leveraging unlabeled streaming data containing both normal and abnormal samples simultaneously. Since the data stream has no ground truth labels, the model needs to filter the anomalous data by detection to avoid contaminating its parameters, which presents a significant challenge when initial samples are insufficient. To tackle this issue, we propose modeling the feature distribution of normal images using a Neural Gas network, which offers the flexibility to adapt the topology structure to identify outliers in the data flow. In order to achieve improved performance with limited training samples, we employ multi-scale feature embedding extracted from a CNN pre-trained on ImageNet to obtain a robust representation. Furthermore, we introduce an algorithm that can incrementally update parameters without the need to store previous samples. Comprehensive experimental results demonstrate that our method can achieve substantial performance under the FOADS setting, while ensuring that the time complexity remains within an acceptable range on MVTec AD and BTAD datasets.

A Benchmark for Morphological Segmentation in Uyghur and Kazakh

Morphological segmentation and stemming are foundational tasks in natural language processing. They have become effective ways to alleviate data sparsity in agglutinative languages because of the nature of agglutinative language word formation. Uyghur and Kazakh, as typical agglutinative languages, have made significant progress in morphological segmentation and stemming in recent years. However, the evaluation metrics used in previous work are character-level based, which may not comprehensively reflect the performance of models in morphological segmentation or stemming. Moreover, existing methods avoid manual feature extraction, but the model’s ability to learn features is inadequate in complex scenarios, and the correlation between different features has not been considered. Consequently, these models lack representation in complex contexts, affecting their effective generalization in practical scenarios. To address these issues, this paper redefines the morphological-level evaluation metrics: F1-score and accuracy (ACC) for morphological segmentation and stemming tasks. In addition, two models are proposed for morpheme segmentation and stem extraction tasks: supervised model and unsupervised model. The supervised model learns character and contextual features simultaneously, then feature embeddings are input into a Transformer encoder to study the correlation between character and context embeddings. The last layer of the model uses a CRF or softmax layer to determine morphological boundaries. In unsupervised learning, an encoder–decoder structure introduces n-gram correlation assumptions and masked attention mechanisms, enhancing the correlation between characters within n-grams and reducing the impact of characters outside n-grams on boundaries. Finally, comprehensive comparative analyses of the performance of different models are conducted from various points of view. Experimental results demonstrate that: (1) The proposed evaluation method effectively reflects the differences in morphological segmentation and stemming for Uyghur and Kazakh; (2) Learning different features and their correlation can enhance the model’s generalization ability in complex contexts. The proposed models achieve state-of-the-art performance on Uyghur and Kazakh datasets.

Which surrogate insulin resistance indices best predict coronary artery disease? A machine learning approach

BackgroundVarious surrogate markers of insulin resistance have been developed, capable of predicting coronary artery disease (CAD) without the need to detect serum insulin. For accurate prediction, they depend only on glucose and lipid profiles, as well as anthropometric features. However, there is still no agreement on the most suitable one for predicting CAD.MethodsWe followed a cohort of 2,000 individuals, ranging in age from 20 to 74, for a duration of 9.9 years. We utilized multivariate Cox proportional hazard models to investigate the association between TyG-index, TyG-BMI, TyG-WC, TG/HDL, plus METS-IR and the occurrence of CAD. The receiver operating curve (ROC) was employed to compare the predictive efficacy of these indices and their corresponding cutoff values for predicting CAD. We also used three distinct embedded feature selection methods: LASSO, Random Forest feature selection, and the Boruta algorithm, to evaluate and compare surrogate markers of insulin resistance in predicting CAD. In addition, we utilized the ceteris paribus profile on the Random Forest model to illustrate how the model’s predictive performance is affected by variations in individual surrogate markers, while keeping all other factors consistent in a diagram.ResultsThe TyG-index was the only surrogate marker of insulin resistance that demonstrated an association with CAD in fully adjusted model (HR: 2.54, CI: 1.34–4.81). The association was more prominent in females. Moreover, it demonstrated the highest area under the ROC curve (0.67 [0.63–0.7]) in comparison to other surrogate indices for insulin resistance. All feature selection approaches concur that the TyG-index is the most reliable surrogate insulin resistance marker for predicting CAD. Based on the Ceteris paribus profile of Random Forest the predictive ability of the TyG-index increased steadily after 9 with a positive slope, without any decline or leveling off.ConclusionDue to the simplicity of assessing the TyG-index with routine biochemical assays and given that the TyG-index was the most effective surrogate insulin resistance index for predicting CAD based on our results, it seems suitable for inclusion in future CAD prevention strategies.

KaTaGCN: Knowledge-Augmented and Time-Aware Graph Convolutional Network for efficient traffic forecasting

Dynamic spatio-temporal dependencies and temporal patterns in traffic series are critical factors affecting traffic forecasting accuracy. Due to the intrinsic challenges of incorporating explicit, logical knowledge into the implicit black-box learning process of neural networks, only a few methods effectively use prior knowledge to improve the learning of traffic forecasting. To tackle this problem, we introduce a new approach called Knowledge-augmented and Time-aware Graph Convolutional Network, namely KaTaGCN. At its core, we have created a knowledge-augmented module that boosts the diffusion weights between closely related adjacent nodes in graph learning. This is achieved by implementing a new loss function. Then, to learn the periodic implicit relationship between these timestamps and traffic signals, the weights and biases are chosen adaptively to be trained based on the timestamps of each sample. Finally, a gated spatio-temporal mapping module regresses high-dimensional embedded features from spatial and temporal dimensions. KaTaGCN is structured without any attention mechanisms or recurrent neural networks. Extensive experimental results on six real-world public traffic datasets demonstrate that KaTaGCN achieves an average improvement of 4.29% in forecasting performance compared with suboptimal results.

Street-level temperature estimation using graph neural networks: Performance, feature embedding and interpretability

Estimating street-level air temperature is a challenging task due to the highly heterogeneous urban surfaces, canyon-like street morphology, and the diverse physical processes in the built environment. Though pioneering studies have embarked on investigations via data-driven approaches, many questions remain to be answered. In this study, we leveraged an innovative framework and redefined the street-level temperature estimation problem using Graph Neural Networks (GNN) with spatial embedding techniques. The results showed that GNN models are more capable and consistent of estimating street-level temperature among tested locations, benefiting from its unique strength in handling extensive data over unstructured graph topology. In addition, we conducted in-depth analysis of feature importance to enhance the model interpretability. Among the urban features analyzed in this study, the time-variant canopy density and meter-level land use data emerge as crucial factors. Our findings highlight GNN's high potential in capturing the complex dynamics between urban elements and their impacts on microclimate, thus offering valuable insights for comprehensive urban data collection and urban climate modeling in general. Collectively, this study also contributes to urban planning and policy by providing avenues to enhance city resilience against climate change, thereby advancing the agenda for environmental stewardship and urban sustainability.

Data-driven estimation of battery state-of-health with formation features

Accurately estimating the state-of-health (SOH) of a battery is crucial for ensuring battery safe and efficient operation. The lifetime of lithium-ion batteries (LIBs) starts from their manufacture, and the performance of LIBs in the service period is highly related to the formation conditions in the factory. Here, we develop a deep transfer ensemble learning framework with two constructive layers to estimate battery SOH. The primary approach involves a combination of base models, a convolutional neural network to combine electrical features with spatial relationships of thermal and mechanical features from formation to subsequent cycles, and long short-term memory to extract temporal dependencies during cycling. Gaussian process regression (GPR) then handles SOH prediction based on this integrated model. The validation results demonstrate highly accurate capacity estimation, with a lowest root-mean-square error (RMSE) of 1.662% and a mean RMSE of 2.512%. Characterization on retired cells reveals the correlation between embedded formation features and their impact on the structural, morphological, and valence states evolution of electrode material, enabling reliable prediction with the corresponding interplay mechanism. Our work highlights the value of deep learning with comprehensive analysis through the relevant features, and provides guidance for optimizing battery management.

DAJLENet: A neural network based on dual attention and joint learning for explainable heart failure adverse event prediction

Heart Failure (HF) is a common, high-risk, complex cardiovascular disease that is often accompanied by other diseases and has a high mortality rate. The use of electronic health records (EHRs) to predict the risk of adverse events, such as death, intensive care units (ICU) transfer, and readmission, in patients with HF is an important topic in the field of personalized treatment and prognosis. However, the heterogeneity and diversity of patient EHRs make it difficult for traditional machine learning models to adequately mine the hidden features and well train patient representations. In addition, significant sample imbalance and lack of labeled data can limit model performance. To address the above issues, this study proposes a neural network called DAJLENet for predicting adverse events in heart failure with explainability. The model obtains feature-level and visit-level patient representations through feature mining, feature embedding, and dual attention mechanisms. Based on the patient representation, a joint training method of classification task and contrastive learning task is used to learn shared representations to alleviate the sample imbalance problem and improve the model performance. Experimental results on four heart failure prediction tasks show that DAJLENet has an improvement of 1%–3% on various metrics compared to the most advanced models, and an improvement of more than 5% on unbalanced datasets.

Zero‐Shot 3D Pose Estimation of Unseen Object by Two‐step RGB-D Fusion

3D Object pose estimation is a critical task in many real-world applications, e.g., robotic manipulation and augmented reality. Most existing methods focus on estimating the object instances or categories which have been seen in the training phase. However, it is imperative to estimate the pose of unseen objects without re-training the network in real world. Therefore, we proposed a 3D pose estimation method for unseen objects without re-training. Specifically, given the CAD model of the unseen object, a set of template RGB-D images (RGB images and depth images) is rendered at different viewpoints. Then a feature embedding network, named PoseFusion, is designed to extract the scene feature. In this network, RGB-D images are utilized to extract the texture feature and geometric feature, respectively. Afterwards, a cross-modality alignment module is proposed to eliminate the noise in single modality. The aligned texture feature and aligned geometric feature are fused through a geometry guided fusion module. Thus, by PoseFusion, the template RGB-D images generated from the CAD model are abstracted into a set of template scene features, and the query scene features are also embedded from the captured RGB-D images from the unseen object. Finally, the query scene features are matched with the template scene features by calculating the masked local similarity. Then the identity and pose of unseen object are determined by the most similar template. Experiments on LINEMOD and T-LESS datasets demonstrate that our method outperforms other methods and generalizes better to unseen objects. Extensive ablation studies are performed to verify the effectiveness of the PoseFusion.

Graphormer-IR: Graph Transformers Predict Experimental IR Spectra Using Highly Specialized Attention.

Infrared (IR) spectroscopy is an important analytical tool in various chemical and forensic domains and a great deal of effort has gone into developing in silico methods for predicting experimental spectra. A key challenge in this regard is generating highly accurate spectra quickly to enable real-time feedback between computation and experiment. Here, we employ Graphormer, a graph neural network (GNN) transformer, to predict IR spectra using only simplified molecular-input line-entry system (SMILES) strings. Our data set includes 53,528 high-quality spectra, measured in five different experimental media (i.e., phases), for molecules containing the elements H, C, N, O, F, Si, S, P, Cl, Br, and I. When using only atomic numbers for node encodings, Graphormer-IR achieved a mean test spectral information similarity (SISμ) value of 0.8449 ± 0.0012 (n = 5), which surpasses that the current state-of-the-art model Chemprop-IR (SISμ = 0.8409 ± 0.0014, n = 5) with only 36% of the encoded information. Augmenting node embeddings with additional node-level descriptors in learned embeddings generated through a multilayer perceptron improves scores to SISμ = 0.8523 ± 0.0006, a total improvement of 19.7σ (t = 19). These improved scores show how Graphormer-IR excels in capturing long-range interactions like hydrogen bonding, anharmonic peak positions in experimental spectra, and stretching frequencies of uncommon functional groups. Scaling our architecture to 210 attention heads demonstrates specialist-like behavior for distinct IR frequencies that improves model performance. Our model utilizes novel architectures, including a global node for phase encoding, learned node feature embeddings, and a one-dimensional (1D) smoothing convolutional neural network (CNN). Graphormer-IR's innovations underscore its value over traditional message-passing neural networks (MPNNs) due to its expressive embeddings and ability to capture long-range intramolecular relationships.

StreamBoostE: A Hybrid Boosting-Collaborative Filter Scheme for Adaptive User-Item Recommender for Streaming Services

With home entertainment, selecting the perfect movie is a pervasive challenge, amplified by many streaming platforms like Netflix and Amazon Prime. This study advances a movie recommender system with collaborative filtering approach as implemented in Python titled StreamBoostE. We used the user-based and item-based similarity schemes on feature embedding to aid faster model construction and training for the tree-based gradient boosting ensemble. Employing both user- and item-based collaborative filtering with cosine similarity to ease feature embedding, the system assesses movies inter-relations via personalized user interest and preferences as submitted user titles with a focus on movie genre classification. Results shows the ensemble yields a recommender prediction accuracy of 0.9984 with F1 of 0.996. The major contribution of StreamBoostE is in its capability to expedite the movie selection process when integrated using flask API and streamlit for cross-channel integration in web-based platforms. It presents users with a list of top-10 recommended movies by genre similarity. The XGBoost ensemble performed best with the user-/item-based collaborative filtering scheme fused with feature embedding approach as a sampling method. Keywords: Random Forest, SMOTE, credit card fraud detection, feature selection, imbalanced dataset Aims Research Journal Reference Format: Atuduhor, R.R., Okpor, M.D., Yoro, R.E., Odiakaose, C.C., Emordi, F.U., Ojugo, A.A., Ako, R.E., Geteloma, V.O., Ejeh, P.O., Abere, R.A., Ifioko, A.M., & Brizimor, S.E. (2024): StreamBoostE: A Hybrid Boosting-Collaborative Filter Scheme for Adaptive User-Item Recommender for Streaming Services. Advances in Multidisciplinary and Scientific Research Journal Vol. 10. No. 2. Pp 89-106. www.isteams.net/aimsjournal. dx.doi.org/10.22624/AIMS/V10N2P8

Joint Anchor Graph Embedding and Discrete Feature Scoring for Unsupervised Feature Selection.

The success of existing unsupervised feature selection (UFS) methods heavily relies on the assumption that the intrinsic relationships among original high-dimensional (HD) data samples exist in the discriminative low-dimension (LD) subspace. However, previous UFS methods commonly construct pairwise graphs and employ l2,1 -norm regularization to severally preserve the local structure and calculate the score of features, which is computationally complex and easy to get stuck into local optimum, so that those approaches cannot be applied in dealing with large-scale datasets in practice. To overcome this challenge, we propose a novel UFS method, in which a novel anchor graph embedding paradigm is designed to extract the local neighborhood relationships among data samples by reducing the computational complexity of graph construction to be linear in the number of data. Moreover, to improve the optimality of selected features as well as the performance of downstream tasks, we propose a discrete feature scoring mechanism, which imposes orthogonal l2,0 -norm constraints on learned projections, in order to enhance the distinction of feature scores as well as reduce the probability of falling into local optimum. In addition, solving the proposed nonconvex and nonsmooth NP-hard problem is challenging, and we present an efficient optimization algorithm to address it and acquire a closed-form solution of the transformation matrix. Extensive experiments demonstrate the effectiveness and efficiency of the proposed UFS by comparison with several state-of-the-art approaches to clustering and image segmentation tasks.

A New Framework of Collaborative Learning for Adaptive Metric Distillation.

This article presents a new adaptive metric distillation approach that can significantly improve the student networks' backbone features, along with better classification results. Previous knowledge distillation (KD) methods usually focus on transferring the knowledge across the classifier logits or feature structure, ignoring the excessive sample relations in the feature space. We demonstrated that such a design greatly limits performance, especially for the retrieval task. The proposed collaborative adaptive metric distillation (CAMD) has three main advantages: 1) the optimization focuses on optimizing the relationship between key pairs by introducing the hard mining strategy into the distillation framework; 2) it provides an adaptive metric distillation that can explicitly optimize the student feature embeddings by applying the relation in the teacher embeddings as supervision; and 3) it employs a collaborative scheme for effective knowledge aggregation. Extensive experiments demonstrated that our approach sets a new state-of-the-art in both the classification and retrieval tasks, outperforming other cutting-edge distillers under various settings.

A Fusion Learning Model Based on Deep Learning for Single-Cell RNA Sequencing Data Clustering.

Single-cell RNA sequencing (scRNA-seq) technology provides a means for studying biology from a cellular perspective. The fundamental goal of scRNA-seq data analysis is to discriminate single-cell types using unsupervised clustering. Few single-cell clustering algorithms have taken into account both deep and surface information, despite the recent slew of suggestions. Consequently, this article constructs a fusion learning framework based on deep learning, namely scGASI. For learning a clustering similarity matrix, scGASI integrates data affinity recovery and deep feature embedding in a unified scheme based on various top feature sets. Next, scGASI learns the low-dimensional latent representation underlying the data using a graph autoencoder to mine the hidden information residing in the data. To efficiently merge the surface information from raw area and the deeper potential information from underlying area, we then construct a fusion learning model based on self-expression. scGASI uses this fusion learning model to learn the similarity matrix of an individual feature set as well as the clustering similarity matrix of all feature sets. Lastly, gene marker identification, visualization, and clustering are accomplished using the clustering similarity matrix. Extensive verification on actual data sets demonstrates that scGASI outperforms many widely used clustering techniques in terms of clustering accuracy.

MTJNet: Multi-task joint learning network for advancing medicinal plant and leaf classification

Identifying and classifying medicinal plants is crucial for pharmaceutical research, traditional medicine, and biodiversity conservation. However, manual classification is time-consuming, error-prone, and requires expert knowledge. This problem statement highlights the need for an automated, efficient, and accurate method by which to distinguish between different species and ensure the quality and safety of medicinal plant leaves. The state-of-the-art methods employ only leaf or plant images for the classification; however, these methods work in particular scenarios (only leaf or plant). In this study, we first integrate the dataset of medicinal plants with corresponding leaves to develop a generalizable model and then apply pre-processing techniques (using vein morphometric features and Sobel edge detection) to highlight the significant characteristics in the images. Then, we propose a novel multi-task joint learning network (MTJNet) for classifying medicinal plant leaves. The proposed MTJNet includes local and global feature extractors that retrieve prominent features for robust classification. Further, we blend the features extracted from the local and global networks to enhance the contextual features. Last, we assign the embedded features to dense layers for additional feature extraction and classification. To evaluate the proposed MTJNet, we employ an Indian medicinal leaf dataset and achieve a precision of 99.60%, a recall of 99.62%, an accuracy of 99.71%, and an F1 score of 99.58%. The experimental results show that the MTJNet statistically defeats prevalent models, thereby proving its effectiveness for medical and industrial applications.

Reduced Order Modeling of System by Dynamic Modal Decom-Position with Fractal Dimension Feature Embedding

Reduced Order Modeling of System by Dynamic Modal Decom-Position with Fractal Dimension Feature Embedding

A Cross-Stage Partial Network and a Cross-Attention-Based Transformer for an Electrocardiogram-Based Cardiovascular Disease Decision System.

Cardiovascular disease (CVD) is one of the leading causes of death globally. Currently, clinical diagnosis of CVD primarily relies on electrocardiograms (ECG), which are relatively easier to identify compared to other diagnostic methods. However, ensuring the accuracy of ECG readings requires specialized training for healthcare professionals. Therefore, developing a CVD diagnostic system based on ECGs can provide preliminary diagnostic results, effectively reducing the workload of healthcare staff and enhancing the accuracy of CVD diagnosis. In this study, a deep neural network with a cross-stage partial network and a cross-attention-based transformer is used to develop an ECG-based CVD decision system. To accurately represent the characteristics of ECG, the cross-stage partial network is employed to extract embedding features. This network can effectively capture and leverage partial information from different stages, enhancing the feature extraction process. To effectively distill the embedding features, a cross-attention-based transformer model, known for its robust scalability that enables it to process data sequences with different lengths and complexities, is employed to extract meaningful embedding features, resulting in more accurate outcomes. The experimental results showed that the challenge scoring metric of the proposed approach is 0.6112, which outperforms others. Therefore, the proposed ECG-based CVD decision system is useful for clinical diagnosis.

GEUKE: A geographic entities uniformly explicit knowledge embedding model

AbstractKnowledge embedding for geographic knowledge graphs can effectively improve computational efficiency and provide support for knowledge reasoning, knowledge answering and other applications of knowledge graphs. To maintain a more comprehensive understanding of spatial features through knowledge embedding, it is crucial to integrate the representation and computation of various entity types, encompassing points, lines, and polygons. This article proposes a geographic entities uniformly explicit knowledge embedding model (GEUKE). In GEUKE, spatial data of point, line, and polygon‐type geographic entities are expressed in the form of subgraphs, and space embedding is generated using a SubGNN‐based uniform spatial feature encoder. GEUKE improves the energy function in TransE to train spatial feature‐based embedding and structural‐based embedding of geographic entities into a unified vector space. Experimental results show that GEUKE has higher performance than TransE, TransH, TransD, and TransE‐GDR on link prediction and triple classification task. Within the spatial feature embedding process, GEUKE effectively preserves the inherent features of entities, encompassing location, neighborhood, and structural attributes, while simultaneously ensuring a coherent spatial data representation across all three entity types: points, lines, and polygons. By maintaining the spatial features of geographic entities and their interrelations, this capability unleashes the full potential of applications such as knowledge reasoning and geospatial question answering in a manner that is conducive to diverse geospatial scenarios.

A feature-aware long-short interest evolution network for sequential recommendation

Recommendation systems are an effective solution to deal with information overload, particularly in the e-commerce sector, in which sequential recommendation is extensively utilized. Sequential recommendations aim to acquire users’ interests and provide accurate recommendations by analyzing users’ historical interaction sequences. To improve recommendation performance, it is vital to take into account the long- and short-term interests of users. Despite significant advancements in this domain, some issues need to be addressed. Conventional sequential recommendation models typically express each item with a uniform embedding, ignoring evolutionary patterns among item attributes, such as category, brand, and price. Moreover, these models often model users’ long- and short-term interests independently, failing to adequately address the issues of interest drift and short-term interest evolution. This study proposes a new model, the Feature-aware Long-Short Interest Evolution Network (FLSIE), to address the above-mentioned issues. Specifically, the model uses explicit feature embedding to represent item attribute information and employs a two-dimensional (2D) attention mechanism to distinguish the significance of individual features in a specific item and the relevance of each item in the interaction sequence. Furthermore, to avoid the issue of interest drift, the model employs a long-term interest guidance mechanism to enhance the representation of short-term interest and adopts a gated recurrent unit with attentional update gate to model the dynamic evolution of users’ short-term interest. Experimental results indicate that our presented model outperforms existing methods on three real-world datasets.

Universal Local Attractors on Graphs

Being able to express broad families of equivariant or invariant attributed graph functions is a popular measuring stick of whether graph neural networks should be employed in practical applications. However, it is equally important to find deep local minima of losses (i.e., produce outputs with much smaller loss values compared to other minima), even when architectures cannot express global minima. In this work we introduce the architectural property of attracting optimization trajectories to local minima as a means of achieving smaller loss values. We take first steps in satisfying this property for losses defined over attributed undirected unweighted graphs with an architecture called universal local attractor (ULA). This refines each dimension of end-to-end-trained node feature embeddings based on graph structure to track the optimization trajectories of losses satisfying some mild conditions. The refined dimensions are then linearly pooled to create predictions. We experiment on 11 tasks, from node classification to clique detection, on which ULA is comparable with or outperforms popular alternatives of similar or greater theoretical expressive power.