graph neural networks

Multi-Agent Reinforcement Learning for Dynamic Inventory Rebalancing and Last-Mile Fulfillment Under Supply Chain Disruptions

Experiments show that prediction must be coupled with autonomous optimization to deliver practical resilience, and show that the learned policy sustains service levels, shortens recovery time, and reduces total disruption cost relative to base-stock and single-agent baselines.

Imitation Learning for Connection-Tableau Construction

An automated theorem prover builds a proof step by step, choosing at each point what to add and what to remove, and learns policies that solve up to 46% more problems than leanCoP, and reach proofs in an order of magnitude fewer steps.

Feasibility Study on Inferring Construction Workers’ Roles and Cooperation Status Using Graph Neural Networks

CrystalGRW: generative modeling of crystal structures with targeted crystallographic properties via geodesic random walks

CrystalGRW is introduced, a diffusion-based generative model on Riemannian manifolds that proposes candidate crystal configurations in stable phases, validated through density functional theory calculations, thereby accelerating materials discovery and inverse design.

基于多模态信息融合的神经符号推理引擎

This paper introduces a novel neuro-symbolic reasoning engine designed to tackle complex problems by integrating the strengths of deep learning and symbolic reasoning. The core concept is to fuse multi-modal information – including images, text, and audio – to create a richer understanding of the problem domain. This understanding is then leveraged through a neuro-symbolic architecture that performs logical inference and knowledge representation. Specifically, the engine utilizes convolutional neural networks (CNNs) for feature extraction from image and audio data, and recurrent neural networks (RNNs) combined with graph neural networks (GNNs) for processing textual information. The fused features are then fed into a symbolic reasoning module, enabling the engine to derive conclusions based on logical rules and existing knowledge. Experimental results (simulated due to the absence of physical experimentation) demonstrate the potential of this approach for improved problem-solving accuracy compared to traditional methods relying solely on either deep learning or symbolic reasoning. The engine's architecture is designed for modularity and extensibility, allowing for the integration of new data modalities and reasoning rules. Future research directions include exploring different fusion techniques and developing more sophisticated symbolic reasoning methods. ---

Enhanced air quality prediction using probabilistic spiking neural networks with multi-aspect graph attention mechanism

基于自适应的神经网络的图论优化

This paper presents a novel graph theory-based optimization algorithm for neural network architecture design, leveraging the inherent properties of graph structures to enhance network performance. Traditional neural network architectures often rely on fixed configurations, while this approach dynamically adjusts network parameters to optimize performance. We introduce a new algorithm that employs adaptive parameter adjustment to optimize neural network structure, resulting in improved training efficiency and enhanced generalization capabilities. This work offers a significant advancement in neural network design by effectively utilizing graph theory principles.

##基于可扩展图神经网络的软件漏洞自动检测

This paper presents a novel approach to automated software vulnerability detection using scalable graph neural networks (GNNs). Traditional vulnerability detection methods often rely on manual code review or signature-based approaches, which are labor-intensive, error-prone, and struggle to identify novel vulnerabilities. Our proposed method leverages the power of GNNs to learn complex relationships within software code represented as graphs, thereby identifying patterns indicative of vulnerabilities. The core of our approach lies in constructing a graph representation of the code, where nodes represent code elements (e.g., functions, variables, statements) and edges represent relationships between them (e.g., data flow, control flow). The GNN then learns embeddings for these nodes, capturing contextual information and vulnerability patterns. We demonstrate the effectiveness of this approach in detecting various types of software vulnerabilities. The results indicate a significant improvement in detection accuracy and efficiency compared to traditional methods. The scalability of the GNN architecture allows it to handle large and complex codebases effectively. This work contributes to the development of automated and intelligent software security solutions.

Graph-Based Representation Learning for Sensor Networks with Dynamic Topology

This paper introduces a novel graph embedding approach specifically designed for sensor networks exhibiting dynamic topologies. Traditional graph embedding techniques often rely on static graph structures, rendering them ineffective in environments where network connections are constantly changing due to node failures, mobility, or environmental factors. Our method, termed Dynamic Graph Embedding (DGE), addresses this limitation by continuously learning node representations, taking into account evolving network connectivity and temporal dependencies. The core idea is to construct a graph representation that reflects the current state of the sensor network, incorporating node state information and leveraging temporal information to refine node embeddings over time. The proposed DGE framework employs a recurrent neural network (RNN) architecture to capture these temporal dependencies and adapt node representations accordingly. We formulate the learning process as an optimization problem, minimizing a loss function that combines node similarity constraints and temporal consistency measures. This approach enables robust and adaptable graph-based representation learning, crucial for applications such as anomaly detection, localization, and data aggregation in dynamic sensor networks. The presented method offers improved performance compared to static graph embedding techniques, particularly in scenarios with significant topological changes.

Title: Dynamic Probabilistic Graphs via High-Dimensional Neural Networks: Semantic Reasoning and Knowledge Graph Construction in Natural Language Processing

This paper introduces a novel dynamic probabilistic graph model, leveraging high-dimensional neural networks, for natural language processing. We propose a mechanism that dynamically adjusts the network parameters to enhance both semantic reasoning and knowledge graph construction capabilities. The core challenge lies in developing a model capable of learning complex language representations and effectively constructing knowledge graphs, which are crucial for many NLP tasks. The study investigates the effectiveness of this approach through extensive experimentation on various semantic reasoning and knowledge graph tasks. The results demonstrate a significant improvement in these tasks, highlighting the potential of this dynamic probabilistic graph model as a foundation for advanced NLP applications.

基于图神经网络的程序合成与优化

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.

Decentralized Learning with Proof-of-Work Consensus for Graph Neural Networks

This paper introduces a novel decentralized learning framework for Graph Neural Networks (GNNs) leveraging a Proof-of-Work (PoW) consensus mechanism. Traditional GNN training relies heavily on centralized servers, leading to vulnerabilities regarding single points of failure and hindering scalability. Our approach addresses these limitations by establishing a distributed, trustless learning environment where nodes compete to solve computationally demanding problems related to the GNN's training. The solution to these problems is then verified by the network, ensuring consensus and enabling decentralized learning. We demonstrate the feasibility and potential benefits of this paradigm, outlining the core mechanism and highlighting its advantages over centralized approaches. The core claim is that decentralized consensus mechanisms, particularly those based on PoW, provide a robust solution to the scalability and reliability issues inherent in traditional GNN training. This work explores the theoretical foundations and presents a conceptual framework for a PoW-based decentralized GNN learning system.