graph neural networks

Congestion Propagation Identification and Prediction Using Self Attention–Based Diffusion Convolutional Approach

A propagation probability matrix is utilizes to identify congestion propagation patterns and finds traffic behavior over 24 h, revealing critical insights into congestion trends in a selected road network and proposing a novel self attention–based diffusion convolutional network (SADCN) that effectively predicts traffic congestion propagation.

Dynamic Semantic Topology Learning and Program Reasoning

This paper introduces a novel approach to program understanding and optimization, termed Dynamic Semantic Topology Learning and Program Reasoning. The core claim is that program behavior is not solely determined by its source code, but significantly influenced by the dynamic semantic relationship network within its runtime environment, which evolves during execution. We propose a mechanism utilizing Graph Neural Networks (GNNs) to construct and continuously update a program semantic topology. Nodes in this topology represent program elements such as functions, variables, and data structures, while edges represent their semantic dependencies – including call relationships, data flow relationships, and control flow relationships. A reinforcement learning-based GNN learning algorithm leverages feedback signals from program execution (e.g., results, errors, resource consumption) to adjust the topology, enhancing its accuracy in reflecting program behavior. This dynamic approach overcomes limitations of static semantic representations like Abstract Syntax Trees (ASTs) and Control Flow Graphs (CFGs) by capturing the complex, runtime-dependent relationships, ultimately leading to improved program understanding and optimization. The methodology offers a new paradigm for analyzing and reasoning about program logic, particularly in dynamic and complex systems.

Based on Multi-modal Data Fusion for Subconscious Decision Simulation

This paper presents a novel approach to simulating subconscious decision-making processes by leveraging multi-modal data fusion. The core idea is to construct a computational model capable of mirroring the complexities of human subconscious decision-making, moving beyond traditional behavioral analysis. We employ a graph neural network (GNN) architecture for robust multi-modal data integration, transforming diverse data streams – including visual, auditory, and tactile information – into a unified representation. This representation is then utilized within a reinforcement learning framework to simulate the subconscious decision-making process, explicitly modeling the interactive effects between different modalities. The resulting model provides a deeper understanding of how individuals make decisions without conscious awareness, offering potential applications in fields such as robotics, human-computer interaction, and cognitive modeling. The key innovation lies in the comprehensive incorporation of multi-modal interactions, providing a more accurate representation of the human subconscious than existing approaches. We define the following key equations to represent the core processes within the model: Let *xi* represent the input vector for modality *i*, where *i* ∈ {V, A, T}, representing Visual, Auditory, and Tactile modalities, respectively. The dimensionality of each *xi* is denoted as *di*. The multi-modal fusion process can be expressed as: * *xfused* = FusionNetwork(*xV*, *xA*, *xT*) Where *xfused* is the fused representation and FusionNetwork is the graph neural network. The reinforcement learning agent's decision-making process is governed by the following equation: * *ai* = argmaxj [Q( *xfused*, *aj* ) + β * R( *xfused*, *aj*)] Where *ai* is the action taken, *Q* is the Q-function estimating the expected reward, *R* is the reward function, and β is a weighting factor. The model's training objective can be formalized as: Minimize Eτ [ Σt=0T γt *R( *xfused*, *at* )] Where τ is a trajectory, *R* is the reward function, γ is the discount factor, and T is the time horizon.

Physics-structured graph neural networks for interpretable fault detection in semiconductor manufacturing processes

Integrating ENSO dynamics into physics-informed Graph Neural Networks for extreme evapotranspiration prediction under climate change

A physics-informed wavelet-based multi-scale graph neural network for aero-engine remaining useful life prediction

Parallel Dual-Level Attention-Enhanced Semantic Segmentation of 3D Point Clouds in Tunnel Construction

Validated on real tunnel data, PDAPointNet++ outperforms models such as PointNet, PointNet++, PointMLP, and dynamic graph convolutional neural network (DGCNN) and the impact of different WDFL balancing coefficients on model performance is discussed.

Graph neural network enhanced large language model for commonsense question answering

Bacteria tracking and life cycle state classification using graph neural networks and pretrained vision transformers.

This work proposes a unified framework based on Graph Neural Networks for simultaneous bacteria tracking, division detection, and life cycle state classification and introduces a network-based approach for segmentation error correction using division and multi-frame correspondence predictions.

PROTEUS: A 40 nm Programmable General-Purpose Digital Compute-In-Memory Accelerator With eNVM and Hierarchical ISA for Versatile Edge AI

We present PROTEUS, an 18 mm2 programmable general-purpose digital compute-in-memory (GP-DCIM) accelerator integrating 4 Mb resistive random access memory (RRAM) and 2.6 Mb tensor static random access memory (SRAM) with a 32-bit hierarchical DCIM instruction set architecture (ISA). PROTEUS features fine-grained 1-D matrix tiling and a reconfigurable DCIM datapath/pipeline for near-100% memory utilization, supporting INT8/INT16/FP8/FP16 DCIM computations. PROTEUS unifies SRAM/RRAM dataflows and embeds nonvolatile micro-programs in RRAM to enable rapid switching among prestored kernels without incurring off-chip instruction feeds or RRAM rewrites. Fabricated in 40 nm ultra-low power (ULP) CMOS with foundry RRAM, PROTEUS delivers 702 GOPS throughput, 6.4 TOPS/W energy efficiency, and 0.039 TOPS/mm2 compute density. It is validated on ResNet-20, BERT-Tiny, MobileViT, GraphSAGE, and Vision Mamba, demonstrating versatility across CNN, Transformer, hybrid CNN-Transformer, graph neural network (GNN), and state-space model (SSM) workloads.

MADGCN: A Meteorology-Aware Spatio-Temporal Graph Convolution Network for Long-Term Air Pollution Forecasting

Air quality forecasting has attracted increasing attention as global air pollution worsens. Spatiotemporal graph neural networks have become a leading paradigm, thanks to their ability to capture complex spatial and temporal dynamics in Air Quality Index (AQI) data. However, existing methods remain limited by weak modeling of long-range temporal dependencies and insufficient integration of meteorological factors. Building on a publicly available nationwide air quality dataset spanning eight years, we propose MADGCN, a Meteorology-Aware Decoupled Spatio-Temporal Convolutional Network that jointly addresses long-horizon temporal modeling and meteorological context fusion. MADGCN includes a dynamic causality discovery module grounded in Granger causality, which captures time-varying causal relationships between meteorological conditions and AQI dynamics. The inferred causal structures further guide a causal graph convolution module and a PatchMixer module, enabling effective spatial interaction modeling and multiscale temporal dependency learning. Extensive experiments against 16 strong baselines show that MADGCN achieves competitive performance for long-horizon air pollution forecasting and generalizes well under high-pollution regimes..

MOSAIC: Multigranularity OOD Detection for IoT Networks via Self-Aligned In-Distribution Consistency

Out-of-distribution (OOD) anomalies pose serious threats to the reliability and security of Internet of Things (IoT) systems. As graph neural networks (GNNs) have become a dominant framework for modeling the relational structures inherent in IoT networks, detecting OOD nodes on graphs has emerged as an essential requirement for trustworthy IoT deployment. For graph node-level OOD detection, current leading methods predominantly follow the OOD-exposure paradigm, which leverages real or synthesized OOD samples to explicitly separate in-distribution (ID) and OOD scores during training. However, this paradigm suffers from two fundamental limitations: 1) real OOD samples are inherently difficult to collect, as novel anomaly patterns in evolving IoT environments are unpredictable before deployment and 2) synthesized pseudo-OOD samples inevitably deviate from true OOD samples, introducing a distributional mismatch that undermines detection reliability. Moreover, both routes rely on regularization hyperparameters that cannot be validated without OOD data. Together, these limitations render OOD-exposure methods illsuited for real-world IoT deployments. To address these issues, we propose multigranularity OOD detection via self-aligned ID consistency (MOSAIC), a graph node-level OOD detection framework that avoids reliance on OOD samples and instead characterizes the ID distribution itself from multiple complementary granularities. MOSAIC evaluates each node from three perspectives: macrolevel global deviation via cosine distance to a distance-weighted ID centroid, mesolevel class-aware boundary deviation via minimum Euclidean distance to class-specific centroids, and microlevel representational stability via feature-masked embedding consistency. A homoscedastic uncertainty framework further balances the training objectives automatically, eliminating the manually tuned regularization hyperparameters required by OOD-exposure methods. Extensive experiments on five social-IoT proxy benchmarks and a bitcoin transaction graph demonstrate that MOSAIC matches or surpasses OOD-exposure-based methods while requiring no OOD data during training, offering a practical solution for open and evolving graph-based IoT systems. The codes are available at https://github.com/Brucesustech/MOSAIC