Graph Architecture Research Articles

Multitarget prediction and optimization of pure electric vehicle tire/road airborne noise sound quality based on a knowledge- and data-driven method

With the promotion of pure electric vehicles (PEVs), the overall interior noise level has been gradually reduced. Tire/road noise is increasingly becoming noticeable in PEVs and represents a primary concern for passengers. Vehicle acoustic packages are crucial for suppressing tire/road noise, and numerous studies have focused on improving the acoustic package performance and sound quality of tire/road noise. However, the prediction and optimization of tire/road acoustic comfort have two deficiencies: (1) The characteristics and transfer paths of tire/road noise are complex. Using knowledge-driven methods (such as simulation models) and data-driven methods (such as neural networks) for analysis has difficulty in accuracy parameter acquisition and unexplainable models. (2) The sound quality of tire/road noise is multidimensional. In the development of multitarget prediction models, when multiple targets contradict each other, prediction bias may result from wide differences among targets in the gradient value of the loss function in training. Therefore, in this paper, a knowledge- and data-driven method is proposed, which introduces the knowledge graph technique to develop a vehicle tire/road noise knowledge graph architecture and uses an improved residual network to drive reasoning in the domain knowledge graph. In addition, a multitarget prediction method based on the adaptive balanced learning mechanism for a residual network is proposed, which uses the dynamic weighted average method to adaptively adjust the loss weight of each target according to the convergence speed and learning difficulty of each target. The proposed dual-drive method is applied to predict and optimize the performance of vehicle acoustic package and to further improve the multiple sound quality metrics of tire/road noise. In the experimental validation, the proposed method outperforms the traditional prediction and optimization methods in effectiveness and robustness.

FAS: assessing the similarity between proteins using multi-layered feature architectures.

Protein sequence comparison is a fundamental element in the bioinformatics toolkit. When sequences are annotated with features such as functional domains, transmembrane domains, low complexity regions or secondary structure elements, the resulting feature architectures allow better informed comparisons. However, many existing schemes for scoring architecture similarities cannot cope with features arising from multiple annotation sources. Those that do fall short in the resolution of overlapping and redundant feature annotations. Here, we introduce FAS, a scoring method that integrates features from multiple annotation sources in a directed acyclic architecture graph. Redundancies are resolved as part of the architecture comparison by finding the paths through the graphs that maximize the pair-wise architecture similarity. In a large-scale evaluation on more than 10000 human-yeast ortholog pairs, architecture similarities assessed with FAS are consistently more plausible than those obtained using e-values to resolve overlaps or leaving overlaps unresolved. Three case studies demonstrate the utility of FAS on architecture comparison tasks: benchmarking of orthology assignment software, identification of functionally diverged orthologs, and diagnosing protein architecture changes stemming from faulty gene predictions. With the help of FAS, feature architecture comparisons can now be routinely integrated into these and many other applications. FAS is available as python package: https://pypi.org/project/greedyFAS/.

Detection of false data injection attacks on power systems using graph edge-conditioned convolutional networks

State estimation plays a vital role in the stable operation of modern power systems, but it is vulnerable to cyber attacks. False data injection attacks (FDIA), one of the most common cyber attacks, can tamper with measurement data and bypass the bad data detection (BDD) mechanism, leading to incorrect results of power system state estimation (PSSE). This paper presents a detection framework of FDIA for PSSE based on graph edge-conditioned convolutional networks (GECCN), which use topology information, node features and edge features. Through deep graph architecture, the correlation of sample data is effectively mined to establish the mapping relationship between the estimated values of measurements and the actual states of power systems. In addition, the edge-conditioned convolution operation allows processing data sets with different graph structures. Case studies are undertaken on the IEEE 14-bus system under different attack intensities and degrees to evaluate the performance of GECCN. Simulation results show that GECCN has better detection performance than convolutional neural networks, deep neural networks and support vector machine. Moreover, the satisfactory detection performance obtained with the data sets of the IEEE 14-bus, 30-bus and 118-bus systems verifies the effective scalability of GECCN.

Knowledge Graph-Based Explainable Artificial Intelligence for Business Process Analysis

For critical operational decisions (e.g. consulting services), explanations and interpretable results of powerful Artificial Intelligence (AI) systems are becoming increasingly important. Knowledge graphs possess a semantic model that integrates heterogeneous information sources and represents knowledge elements in a machine-readable form. The integration of knowledge graphs and machine learning methods represents a new form of hybrid intelligent systems that benefit from each other’s strengths. Our research aims at an explainable system with a specific knowledge graph architecture that generates human-understandable results even when no suitable domain experts are available. Against this background, the interpretability of a knowledge graph-based explainable AI approach for business process analysis is focused. We design a framework of interpretation, show how interpretable models are generated by a single case study and evaluate the applicability of our approach by different expert interviews. Result paths on weaknesses and improvement measures related to a business process are used to produce stochastic decision trees, which improve the interpretability of results. This can lead to interesting consulting self-services for clients or be applied as a device for accelerating classical consulting projects.

An herbal drug combination identified by knowledge graph alleviates the clinical symptoms of plasma cell mastitis patients: a nonrandomized controlled trial.

Plasma cell mastitis (PCM) is a nonbacterial breast inflammation with severe and intense clinical manifestation, yet treatment methods for PCM are still rather limited. Although the mechanism of PCM remains unclear, mounting evidence suggests that the dysregulation of immune system is closely associated with the pathogenesis of PCM. Drug combinations or combination therapy could exert improved efficacy and reduced toxicity by hitting multiple discrete cellular targets. We have developed a knowledge graph architecture toward immunotherapy and systematic immunity that consists of herbal drug-target interactions with a novel scoring system to select drug combinations based on target-hitting rates and phenotype relativeness. To this end, we employed this knowledge graph to identify an herbal drug combination for PCM and we subsequently evaluated the efficacy of the herbal drug combination in clinical trial. Our clinical data suggests that the herbal drug combination could significantly reduce the serum level of various inflammatory cytokines, downregulate serum IgA and IgG level, reduce the recurrence rate, and reverse the clinical symptoms of PCM patients with improvements in general health status. In summary, we reported that an herbal drug combination identified by knowledge graph can alleviate the clinical symptoms of PCM patients. We demonstrated that the herbal drug combination holds great promise as an effective remedy for PCM, acting through the regulation of immunoinflammatory pathways and improvement of systematic immune level. In particular, the herbal drug combination could significantly reduce the recurrence rate of PCM, a major obstacle to PCM treatment. Our data suggests that the herbal drug combination is expected to feature prominently in future PCM treatment. C. Liu's lab was supported by grants from the Public Health Science and Technology Project of Shenyang (grant: 22-321-32-18); Y. Yang's laboratory was supported by the National Natural Science Foundation of China (grant: 81874301), the Fundamental Research Funds for Central University (grant: DUT22YG122), and the Key Research project of 'be Recruited and be in Command' in Liaoning Province (2021JH1/10400050). NCT05530226.

Blume–Emery–Griffiths model on random graphs

The Blume–Emery–Griffiths model with a random crystal field is studied in a random graph architecture, in which the average connectivity is a controllable parameter. The disordered average over the graph realizations is treated by replica symmetry formalism of order parameter functions. A self consistent equation for the distribution of local fields is derived, and numerically solved by a population dynamics algorithm. The results show that the average connectivity amounts to changes in the topology of the phase diagrams. Phase diagrams for representative values of the model parameters are compared with those obtained for fully connected mean field and renormalization group approaches.

Graph Convolutional Neural Networks for Histologic Classification of Pancreatic Cancer.

Pancreatic ductal adenocarcinoma has some of the worst prognostic outcomes among various cancer types. Detection of histologic patterns of pancreatic tumors is essential to predict prognosis and decide the treatment for patients. This histologic classification can have a large degree of variability even among expert pathologists. To detect aggressive adenocarcinoma and less aggressive pancreatic tumors from nonneoplasm cases using a graph convolutional network-based deep learning model. Our model uses a convolutional neural network to extract detailed information from every small region in a whole slide image. Then, we use a graph architecture to aggregate the extracted features from these regions and their positional information to capture the whole slide-level structure and make the final prediction. We evaluated our model on an independent test set and achieved an F1 score of 0.85 for detecting neoplastic cells and ductal adenocarcinoma, significantly outperforming other baseline methods. If validated in prospective studies, this approach has a great potential to assist pathologists in identifying adenocarcinoma and other types of pancreatic tumors in clinical settings.

Learning the dynamics of particle-based systems with Lagrangian graph neural networks

Physical systems are commonly represented as a combination of particles, the individual dynamics of which govern the system dynamics. However, traditional approaches require the knowledge of several abstract quantities such as the energy or force to infer the dynamics of these particles. Here, we present a framework, namely, Lagrangian graph neural network (LGnn), that provides a strong inductive bias to learn the Lagrangian of a particle-based system directly from the trajectory. We test our approach on challenging systems with constraints and drag—LGnn outperforms baselines such as feed-forward Lagrangian neural network (Lnn) with improved performance. We also show the zero-shot generalizability of the system by simulating systems two orders of magnitude larger than the trained one and also hybrid systems that are unseen by the model, a unique feature. The graph architecture of LGnn significantly simplifies the learning in comparison to Lnn with ∼25 times better performance on ∼20 times smaller amounts of data. Finally, we show the interpretability of LGnn, which directly provides physical insights on drag and constraint forces learned by the model. LGnn can thus provide a fillip toward understanding the dynamics of physical systems purely from observable quantities.

Large-Scale Hyperspectral Image Restoration via a Superpixel Distributed Algorithm Based on Graph Signal Processing

Hyperspectral image (HSI) is often disturbed by various kinds of noise, which brings great challenges to subsequent applications. Many of the existing restoration algorithms do not scale well for HSI with large size. This paper proposes a novel mixed-noise removal method for HSI with large size, by leveraging the superpixel segmentation-based technology and distributed algorithm based on graph signal processing. First, the underlying structure of the HSI is modeled by a two-layer architecture graph. The upper layer, called skeleton graph, is a rough graph constructed by using the modified <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">k</i> -nearest-neighborhood algorithm and its nodes correspond to a series of superpixels formed by HSI segmentation. The skeleton graph can efficiently characterize the inter-correlations between superpixels, while preserving the boundary information and reducing the computational complexity. The lower layer, called detailed graph consisting of a series of local graphs which are constructed to model the similarities between pixels. Second, based on the two-layer graph architecture, the HSI restoration problem is formulated as a series of optimization problems each of which resides on a subgraph. In each optimization problem, a graph Laplacian regularization is defined and incorporated into a low-rank-based model. Third, a novel distributed algorithm is tailored for the restoration problem, by using the information interaction between the nodes of skeleton graph and subgraphs. Numerical experiments conducted on both synthetic and real-world datasets demonstrate the effectiveness of the proposed restoration algorithm compared with existing methods.

Deep learning models of cognitive processes constrained by human brain connectomes

Decoding cognitive processes from recordings of brain activity has been an active topic in neuroscience research for decades. Traditional decoding studies focused on pattern classification in specific regions of interest and averaging brain activity over many trials. Recently, brain decoding with graph neural networks has been shown to scale at fine temporal resolution and on the full brain, achieving state-of-the-art performance on the human connectome project benchmark. The reason behind this success is likely the strong inductive connectome prior that enables the integration of distributed patterns of brain activity. Yet, the nature of such inductive bias is still poorly understood. In this work, we investigate the impact of the inclusion of multiple path lengths (through high-order graph convolution), the homogeneity of brain parcels (graph nodes), and the type of interactions (graph edges). We evaluate the decoding models on a large population of 1200 participants, under 21 different experimental conditions, acquired from the Human Connectome Project database.Our findings reveal that the optimal choice for large-scale cognitive decoding is to propagate neural dynamics within empirical functional connectomes and integrate brain dynamics using high-order graph convolutions. In this setting, the model exhibits high decoding accuracy and robustness against adversarial attacks on the graph architecture, including randomization in functional connectomes and lesions in targeted brain regions and networks. The trained model relies on biologically meaningful features for the prediction of cognitive states and generates task-specific graph representations resembling task-evoked activation maps. These results demonstrate that a full-brain integrative model is critical for the large-scale brain decoding. Our study establishes principles of how to effectively leverage human connectome constraints in deep graph neural networks, providing new avenues to study the neural substrates of human cognition at scale.

Gravity++: A graph-based framework for constructing interactive visualization narratives

Interactive visualizations play a key part in the exploration and analysis of data, and in the creation of visual data stories. This paper describes a new graph-based framework for developing interactive visualizations for creating coherent visual data stories. We have realized our framework in a prototype tool named Gravity++. Gravity++ uses a novel graph architecture for modeling interaction, data navigation, and changes in visual representation to better communicate findings to an audience. The combination of these graph models provides better support and flexibility for designing interactive visualizations, data navigation, and visual data stories. We demonstrate the applicability of this framework by two example usage scenarios. We also report on an evaluation study conducted with representative participants. All participants successfully created meaningful visual data stories with a high level of complexity. Our results also show that Gravity++ is easy to use and supports the understanding and sense-making of the visual data story creation process.

Multi-scale rotation-equivariant graph neural networks for unsteady Eulerian fluid dynamics

The simulation of fluid dynamics, typically by numerically solving partial differential equations, is an essential tool in many areas of science and engineering. However, the high computational cost can limit application in practice and may prohibit exploring large parameter spaces. Recent deep-learning approaches have demonstrated the potential to yield surrogate models for the simulation of fluid dynamics. While such models exhibit lower accuracy in comparison, their low runtime makes them appealing for design-space exploration. We introduce two novel graph neural network (GNN) models, multi-scale (MuS)-GNN and rotation-equivariant (RE) MuS-GNN, for extrapolating the time evolution of the fluid flow. In both models, previous states are processed through multiple coarsening of the graph, which enables faster information propagation through the network and improves the capture and forecast of the system state, particularly in problems encompassing phenomena spanning a range of length scales. Additionally, REMuS-GNN is architecturally equivariant to rotations, which allows the network to learn the underlying physics more efficiently, leading to improved accuracy and generalization. We analyze these models using two canonical fluid models: advection and incompressible fluid dynamics. Our results show that the proposed GNN models can generalize from uniform advection fields to high-gradient fields on complex domains. The multi-scale graph architecture allows for inference of incompressible Navier–Stokes solutions, within a range of Reynolds numbers and design parameters, more effectively than a baseline single-scale GNN. Simulations obtained with MuS-GNN and REMuS-GNN are between two and four orders of magnitude faster than the numerical solutions on which they were trained.

Video as Conditional Graph Hierarchy for Multi-Granular Question Answering

Video question answering requires the models to understand and reason about both the complex video and language data to correctly derive the answers. Existing efforts have been focused on designing sophisticated cross-modal interactions to fuse the information from two modalities, while encoding the video and question holistically as frame and word sequences. Despite their success, these methods are essentially revolving around the sequential nature of video- and question-contents, providing little insight to the problem of question-answering and lacking interpretability as well. In this work, we argue that while video is presented in frame sequence, the visual elements (e.g., objects, actions, activities and events) are not sequential but rather hierarchical in semantic space. To align with the multi-granular essence of linguistic concepts in language queries, we propose to model video as a conditional graph hierarchy which weaves together visual facts of different granularity in a level-wise manner, with the guidance of corresponding textual cues. Despite the simplicity, our extensive experiments demonstrate the superiority of such conditional hierarchical graph architecture, with clear performance improvements over prior methods and also better generalization across different type of questions. Further analyses also demonstrate the model's reliability as it shows meaningful visual-textual evidences for the predicted answers.

How to optimize Recommendation System Performance using Deep Neural Network based Graph Architecture

In this research, would bring a description and comparison of how the Deep learning-based Graph neural Network actually outperforms the other similar recommender system like collaborative filtering, content-based filtering, SVD, Matrix Factorization and few others. This is basically achieved by exposing the correct relation between the objects through a graph architecture and the dependency and inter correlation between them. Would also like to share an in-depth analysis and understanding of how the Graph architecture works and the underlying theories. This could be either a TensorFlow based architecture or a Pytorch based architecture but in this paper will mainly focus on the TensorFlow one for its flexibility and cloud friendly nature for adopting in any framework.&#x0D; &#x0D;

Research on intelligent path planning technology of logistics robots based on Giraph architecture

In the development of modern e-commerce, the logistics robot is an important part of the intelligent warehouse. Especially in large-scale warehouses, the operation efficiency of logistics robots largely depends on the processing ability of patrol efficiency. In order to solve the problems of difficult path planning and complex operation in the logistics robot operating system, based on the analysis of the traditional path algorithm, an improved logistics robot path planning algorithm based on graph architecture, CAA*path planning algorithm, is proposed and used to solve the conflict-free path planning problem of the logistics robot in the intelligent logistics warehouse. The improved CAA*path planning algorithm is compared with the traditional a*algorithm, ant colony algorithm, simulated annealing algorithm, ant colony optimisation algorithm and genetic algorithm. The simulation analysis shows that the improved CAA*path planning algorithm is more efficient in time operation. Under the same number of tasks and the same map specification, the improved CAA*path planning algorithm takes less time and runs more efficiently.

Qubit Mapping Based on Subgraph Isomorphism and Filtered Depth-Limited Search

Mapping logical quantum circuits to Noisy Intermediate-Scale Quantum (NISQ) devices is a challenging problem which has attracted rapidly increasing interests from both quantum and classical computing communities. This article proposes an efficient method by (i) selecting an initial mapping that takes into consideration the similarity between the architecture graph of the given NISQ device and a graph induced by the input logical circuit and (ii) searching, in a filtered and depth-limited way, a most useful <sc xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">swap</small> combination that makes executable as many as possible two-qubit gates in the logical circuit. The proposed circuit transformation algorithm can significantly decrease the number of auxiliary two-qubit gates required to be added to the logical circuit, especially when it has a large number of two-qubit gates. For an extensive benchmark set of 131 circuits and IBM's current premium Q system, viz., IBM Q Tokyo, our algorithm needs, in average, 0.3801 extra two-qubit gates per input two-qubit gate, while the corresponding figures for three state-of-the-art algorithms are 0.4705, 0.8154, and 1.0066, respectively.

A Universal Simulation Framework of Shipborne Inertial Sensors Based on the Ship Motion Model and Robot Operating System

A complete virtual test environment is a powerful tool for Autonomous Surface Vessels (ASVs) research, and the simulation of ship motion and shipborne sensors is one of the prerequisites for constructing such an environment. This paper proposed a universal simulation framework of shipborne inertial sensors. A ship motion model considering environmental disturbances is proposed to simulate the six-degrees-of-freedom motion of ships. The discrete form of the inertial sensor stochastic error model is derived. The inertial measurement data are simulated by adding artificial errors to a simulated motion status. In addition, the ship motion simulation, inertial measurement simulation, and environment simulation nodes are implemented based on the computational graph architecture of the Robot Operating System (ROS). The benefit from the versatility of the ROS messages, the format of simulated inertial measurement is exactly the same as that of real sensors, which provides a research basis for the fusion perception algorithm based on visual–inertial and laser–inertial sensors in the research field of ASVs.

Prelude to Simulations of Loop Quantum Gravity on Adiabatic Quantum Computers

The article addresses the possibility of implementing spin network states, used in the loop quantum gravity approach to Planck scale physics on an adiabatic quantum computer. The discussion focuses on applying currently available technologies and analyzes a concrete example of a D-Wave machine. It is introduced a class of simple spin network states which can be implemented on the Chimera graph architecture of the D-Wave quantum processor. However, extension beyond the currently available quantum processor topologies is required to simulate more sophisticated spin network states. This may inspire new generations of adiabatic quantum computers. A possibility of simulating loop quantum gravity is discussed, and a method of solving a graph non-changing scalar (Hamiltonian) constraint with the use of adiabatic quantum computations is proposed. The presented results establish a basis for the future simulations of Planck scale physics, specifically quantum cosmological configurations, on quantum annealers.

FC-GAGA: Fully Connected Gated Graph Architecture for Spatio-Temporal Traffic Forecasting

Forecasting of multivariate time-series is an important problem that has applications in traffic management, cellular network configuration, and quantitative finance. A special case of the problem arises when there is a graph available that captures the relationships between the time-series. In this paper we propose a novel learning architecture that achieves performance competitive with or better than the best existing algorithms, without requiring knowledge of the graph. The key element of our proposed architecture is the learnable fully connected hard graph gating mechanism that enables the use of the state-of-the-art and highly computationally efficient fully connected time-series forecasting architecture in traffic forecasting applications. Experimental results for two public traffic network datasets illustrate the value of our approach, and ablation studies confirm the importance of each element of the architecture. The code is available here: https://github.com/boreshkinai/fc-gaga.

Calcification detection for intravascular ultrasound image using direct acyclic graph architecture: pre-trained model for 1-channel image

Coronary artery calcification is a calcium buildup within the walls of the arteries. It is considered a predominant marker for coronary artery disease. Thus many approaches have been developed for the automatic detection of calcification. The previous calcification detection was on segmentation of other structures as pre-processing steps or using the fact that the calcification often appears as a bright region. In this paper, an automated system proposed using a deep learning approach to detect the calcification absence and calcification presence in coronary artery IVUS image. A useful advantage of deep learning, compared to other methods is, it uses representations and features directly from the raw data, bypassing the need to manually extract features, a common that required in the traditional machine learning framework. The type of deep learning architecture used is 27 layers of convolutional neural networks (CNNs) using Direct Acyclic Graph. The proposed system used 2175 images and achieved an accuracy of 98.16% for Cartesian coordinate images and 99.08% for Polar Reconstructed Coordinate images.