This paper proposes a novel approach to understanding and predicting complex system dynamics by leveraging the principles of self-organization and multi-scale event propagation. The core idea is to construct a self-organized event graph that mimics the behavior of biological neural networks, specifically focusing on the propagation of events across multiple scales. This is achieved through the design of a hierarchical network structure where different layers process event information at varying scales. An adaptive algorithm then dynamically adjusts node connections and event propagation pathways within the graph based on the observed event dynamics. The resulting event graph provides a powerful tool for capturing the intricate patterns and dependencies inherent in complex systems. Mathematical formulation focuses on event propagation, graph structure evolution, and the adaptive learning process. Key elements include event representation, graph adjacency matrix, propagation rules, and the objective function guiding network adaptation. The system is designed to learn patterns of complex behavior from data, offering a new perspective on modeling dynamic systems. ---
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper introduces a novel approach to compiler optimization decision modeling by leveraging Graph Neural Networks (GNNs). Traditional compiler optimization techniques rely heavily on manually crafted rules and heuristics, often proving inadequate for complex codebases and emerging hardware architectures. This research proposes a probabilistic model built upon GNNs to represent and predict optimal optimization decisions. The model utilizes graph structures to encapsulate the intricate relationships within the compiler optimization process, with nodes representing optimization opportunities and edges signifying dependencies between them. Through neural network learning, the GNN dynamically adapts to the specific characteristics of the code, leading to more intelligent and effective optimization strategies. This work demonstrates the potential of GNNs to overcome the limitations of traditional methods and pave the way for automated, data-driven compiler optimization. The core claim is to utilize GNNs to build a probabilistic model for compiler optimization decisions, achieving intelligent optimization. The core mechanism involves modeling the compiler optimization process as a GNN, using graph nodes and edges to represent optimization factors and enabling learning and prediction through neural networks.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper explores the application of Graph Neural Networks (GNNs) to the fields of program synthesis and optimization. The core idea is to represent program code as graph structures and leverage the power of GNNs to automatically learn the underlying structure and semantics. This approach offers a novel pathway for program development, moving beyond traditional, rule-based methods. We detail a framework where GNNs are trained on codebases, learning representations that can then be utilized for automated code generation or optimization tasks. The presented method aims to capture complex relationships within code, leading to more efficient and potentially novel solutions. The research contributes to a shift towards data-driven program development, utilizing the expressive power of GNNs to tackle previously challenging problems. The key contributions include the formalization of the process and a theoretical understanding of how GNNs can be effectively applied to this domain.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper introduces a novel approach to graph embedding that addresses the limitations of existing methods in capturing the dynamic nature of evolving graphs. Traditional graph embedding techniques often rely on static representations, failing to account for the temporal changes in relationships between nodes. Our proposed method, Temporal Relational Graph Embedding with Dynamic Context (TRGDC), leverages a recurrent neural network to model the dynamic context of a graph, incorporating both relational information and the elapsed time since the last update. The core idea is to continuously adapt node embeddings based on these temporal and relational factors. We demonstrate that TRGDC significantly improves the representation of dynamic graphs compared to static embedding methods, achieving superior performance in downstream graph-related tasks. The key contributions of this work are the integration of temporal modeling with relational graph embedding and the utilization of a recurrent neural network for dynamic context representation.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Reach audiences
Advertise in front of researchers, engineers, and readers.
This paper introduces a novel approach to program optimization leveraging the power of Graph Neural Networks (GNNs). The core idea is to represent program abstractions, such as Control Flow Graphs (CFGs) and Data Dependency Graphs (DDGs), as graphs and utilize GNNs to automatically identify optimization opportunities and generate optimized code. Traditional program optimization techniques often rely on hand-crafted heuristics and domain-specific knowledge. This approach offers a more automated and potentially more effective method by allowing the GNN to learn complex optimization patterns directly from the program's structure. We demonstrate the feasibility and potential of this approach through theoretical analysis and a conceptual framework. The key contribution lies in the application of GNNs to program optimization, providing a fundamentally new paradigm for achieving better code optimization. ---
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
This paper presents a novel approach to graph neural network (GNN) learning, leveraging an adaptive graph neural network (AGNN) to dynamically learn the graph topology and subsequently utilize this topology for data modeling. Traditional GNNs typically rely on static graph structures, limiting their ability to effectively capture complex relationships within the data. This work introduces an AGNN that adapts its network architecture and weights to the specific graph structure, enhancing the model's capacity to represent and extract meaningful information. We demonstrate the effectiveness of this method through several experiments, showcasing improved data modeling and analysis efficiency compared to existing approaches.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Graph Neural Networks (GNNs) have achieved remarkable success in various domains, including social network analysis, recommendation systems, and knowledge graph reasoning. However, a significant limitation of traditional GNNs is their inability to effectively handle evolving graph structures and temporal dependencies. This paper proposes a novel Temporal Graph Neural Network (TGNN) architecture, incorporating a Contextual Drift Model, designed to address this challenge. The TGNN learns and predicts the rate of change in node and edge attributes over time, explicitly modeling the concept of "contextual drift." This information is then integrated into the GNN's message-passing process, allowing the model to adapt dynamically to changing graph dynamics. The core mechanism utilizes a recurrent neural network (RNN) layer to capture the temporal evolution of attributes. The resulting TGNN demonstrates improved performance in scenarios where graph structures and node/edge attributes change over time, offering a more robust and adaptable solution compared to static GNNs. The primary contribution lies in the integration of temporal dynamics and contextual information within a graph neural network framework, paving the way for more sophisticated and reliable graph representations in dynamic environments.
Jincheng Zhang· Zenodo (CERN European Organi...· 0 citations
Power communication networks in virtual power plants (VPPs) play a critical role in enabling reliable information exchange among distributed energy resources, energy storage systems, controllable loads, and dispatch centers. However, their complex topology, strong inter-node dependency, and hidden fault propagation paths make accurate state awareness and fault diagnosis highly challenging for conventional rule-based and shallow learning methods. To address this issue, this paper proposes a graph neural network (GNN)-based state awareness and fault diagnosis method for power communication networks in VPPs. First, a node-link coupled graph representation is constructed by integrating network topology and multi-source operational data, so as to capture the correlations among devices, communication links, and service states. Then, a GNN-based framework is developed to learn discriminative state features through neighborhood aggregation and graph representation learning, enabling the identification of multiple fault types, including link anomalies, node failures, and local congestion. In addition, temporal state evolution is incorporated to improve the model’s ability to characterize dynamic operating conditions and complex disturbances. Experimental results demonstrate that the proposed method outperforms conventional machine learning and mainstream deep learning approaches in terms of state recognition accuracy, fault diagnosis precision, and robustness. The proposed method provides an effective solution for intelligent monitoring, online state awareness, and fault early warning in VPP communication networks.
Tingting Ren, Caixiao Ouyang, Li Yang et al.· 0 citations
Fault diagnosis of rotating machinery is essential for ensuring the reliable and safe operation of industrial equipment. However, imbalanced training data often bias intelligent diagnostic models toward majority classes, resulting in insufficient representation of minority faults. In addition, the high dimensionality and redundancy of vibration signals make it difficult to extract discriminative features and preserve meaningful neighborhood structures, thereby degrading diagnostic performance. To address these challenges, a waveform-prediction augmentation and deep manifold learning model for imbalanced fault diagnosis is proposed for rotating machinery. First, minority-class fault signals are augmented through time-series prediction using a nonlinear autoregressive neural network (NARNN), thereby alleviating the class imbalance at the data level. Subsequently, a deep manifold feature mapping (DMAP) is proposed to extract discriminative fault features. In DMAP, a stacked autoencoder (SAE) is employed to perform preliminary feature extraction on the training samples and obtain deep latent feature representations. The deep features extracted by the SAE are further processed using a dynamic time warping (DTW)-assisted manifold learning method for dimensionality reduction. In this process, DTW is used to calculate pairwise dissimilarities between deep representations and construct the neighborhood graph, thereby better preserving the local neighborhood relationships and topological structure of the deep features and producing more discriminative low-dimensional representations. Finally, a k-nearest neighbor classifier is adopted for fault classification. By integrating data augmentation, deep feature learning, and manifold dimensionality reduction, the proposed model effectively improves the fault diagnosis performance in imbalanced scenarios. Experiments on the bearing and planetary gearbox datasets demonstrate that, under an imbalance ratio of 0.1, the proposed method improves diagnostic accuracy by 10.2 and 1.6 percentage points, respectively. The comparative results further confirm its effectiveness in alleviating the diagnostic difficulties caused by class imbalance.
Yu Tian, Shunsheng Guo, Yibing Li et al.· Sensors· 0 citations
Predicting spatiotemporal patterns is critical for traffic flow forecasting in Intelligent Transportation Systems (ITS), as accurate predictions can significantly enhance traffic management and decision-making. Recent data-driven approaches, particularly Graph Neural Networks (GNNs) integrated with physics-informed partial differential equations (PDEs), have demonstrated strong capability in modeling complex traffic dynamics. However, these models may still produce overconfident yet incorrect predictions, which can be particularly problematic in safety-critical scenarios. Uncertainty quantification (UQ) enhances the reliability of predictions and the practical application of NNs by estimating prediction confidence, not just accuracy. Despite its importance, existing UQ methods in traffic forecasting are primarily developed for purely data-driven models, and the role of physics-informed modeling in uncertainty estimation remains insufficiently understood. In this paper, we bridge this gap by integrating multiple UQ baseline methods with physics-informed modeling to systematically investigate how physical constraints influence uncertainty estimation in traffic forecasting. We further propose a physics-informed loss function that enhances the model's ability to capture physically consistent dynamics and improves the calibration and reliability of uncertainty estimates. In addition, we evaluate robustness under both noisy and adversarial perturbations, showing that our approach yields more stable predictions and uncertainty estimates under distribution shifts. Extensive experiments on real-world traffic datasets demonstrate that our approach improves both prediction accuracy and uncertainty quality, achieving up to 14.8% improvement in short-term and 8.7% in long-term traffic speed prediction errors, while providing better-calibrated and more robust uncertainty estimates.
Tianshu Bao, Dung T Nguyen, Xiaoou Liu et al.· ACM Transactions on Cyber-Ph...· 0 citations
Reproducible computational workflow for the identification and prioritization of putative non-covalent KEAP1-NRF2 protein-protein interaction inhibitors. The workflow integrates structure-based virtual screening, binding-pocket analysis, molecular docking, MM-GBSA rescoring, molecular dynamics simulations, transfer-learned graph neural networks, Random Forest regression with SHAP interpretability, and symbolic regression for computational lead prioritization.
Umair Wahedi· Zenodo (CERN European Organi...· 0 citations
Abstract High-precision 3D reconstruction is essential for assessing external quality and extracting phenotypic parameters of navel oranges. Conventional Structure-from-Motion (SfM) pipelines struggle with the weakly textured characteristics and highly specular surfaces of navel oranges, resulting in incomplete reconstructions, sparse point clouds, and severe visual artifacts. To address these challenges, this study develops a high-fidelity 3D reconstruction framework that integrates deep-learning-based feature matching with 3D Gaussian Splatting (3DGS). For the sparse reconstruction phase, the SuperPoint detector and SuperGlue matcher are employed, leveraging deep feature extraction and graph neural networks to enhance matching robustness on complex surfaces. To address the initialization challenges of 3DGS, a spatial-colorimetric cascade purification strategy—combining pass-through filtering, HSV masking, and statistical filtering—is developed to effectively eliminate noise and provide a high-confidence initialization prior. Subsequently, 3DGS is utilized for dense reconstruction and high-fidelity neural rendering. Experimental results demonstrate that the developed framework significantly outperforms the conventional SfM pipeline in terms of registered image count, point cloud density, and trajectory length. The 3DGS-optimized model exhibits substantial improvements across quantitative metrics (PSNR, SSIM, and LPIPS), effectively reducing specular artifacts and enhancing geometric fidelity. This approach offers an efficient digital modeling solution for weakly textured and highly specular fruits, thereby facilitating phenotypic analysis in smart agriculture.
Assistant Professor Pat Pataranutaporn describes a new interface that lets everyday users glimpse inside an AI's neural network before their chatbot ever says a word.
MIT News · Artificial Intelligence· news.mit.eduJul 6, 2026
PhD student Rachel Sava, winner of the Envisioning the Future of Computing Prize, explores transformative improvements and dystopian risks of neural technology.
