gnn allocation policy

To address the low efficiency and poor generalizability of manual configuration and tuning of Graph Neural Networks (GNNs), this paper proposes a knowledge-guided, LLM-driven automated GNN design framework. Methodologically, it constructs a structured graph learning knowledge base and integrates Retrieval-Augmented Generation (RAG) with a multi-agent cooperative evolutionary mechanism, enabling end-to-end autonomous design and optimization of GNN architectures, hyperparameters, and training strategies by large language models. Its core contribution is the introduction of the first “knowledge–retrieval–evolution” closed-loop paradigm, explicitly incorporating domain-specific knowledge into the AutoML pipeline for GNNs. Extensive experiments across 12 benchmark datasets and three graph learning tasks demonstrate that the proposed method achieves an average performance gain of 12.7% over manually tuned GNNs while reducing configuration time by 90%, significantly enhancing both automation capability and cross-task generalizability.

Complex job scheduling faces challenges including resource constraints, diverse operational constraints, and scarcity of labeled training data. Method: This paper proposes an end-to-end adaptive allocation framework integrating Proximal Policy Optimization (PPO) with graph neural networks (GAT or GraphSAGE). It is the first to jointly model the dynamic job–resource matching process using reinforcement learning (RL) and graph neural networks (GNNs), employing constraint-aware graph-structured environment modeling and sparse reward design to enable unsupervised, real-time optimal decision-making. Contribution/Results: Evaluated on both synthetic and real-world datasets, the framework achieves a 12.7% improvement in task completion rate over conventional heuristic and supervised-learning baselines. It demonstrates significantly enhanced generalization capability and online responsiveness, establishing a scalable, label-free paradigm for constrained scheduling problems.

This work addresses the absence of a systematic taxonomy and unified framework for communication mechanisms in graph neural network (GNN)-driven multi-agent reinforcement learning. To bridge this gap, the paper proposes a general GNN-based communication pipeline and establishes the first structured survey and classification scheme, clearly delineating the core mechanisms and design principles underlying such approaches. By integrating insights from GNNs, multi-agent reinforcement learning, and communication modeling, this study enhances conceptual clarity and accessibility in the field, while also laying a theoretical foundation and offering methodological guidance for future research.

This work addresses the expressive power bottleneck of universal policy learning in classical planning. Standard graph neural networks (GNNs) are limited to C₂-logic expressivity, while 3-GNNs achieve C₃ expressivity at prohibitive computational and memory cost. We propose the parametric relational GNN, R-GNN[t], the first GNN variant to progressively approximate C₃ expressivity via input-side tunable transformations—without architectural redesign. The hyperparameter *t* enables flexible trade-offs between expressivity and efficiency, maintaining only quadratic space complexity *O(n²)*. Our theoretical analysis integrates the *k*-GNN framework with first-order logic expressibility, enhanced by optimized message passing and embedding compression. Experiments on multiple planning benchmarks show that R-GNN[1] significantly outperforms standard R-GNN and Edge Transformer: it reduces training memory to *O(n²)*, accelerates inference, and achieves higher policy generalization accuracy.

Graph neural networks (GNNs) suffer from catastrophic prediction failure under asynchronous inference in distributed multi-agent systems—severely limiting their deployment in resource-constrained, inherently desynchronized settings such as robot swarms and sensor networks. To address this, we first establish the first theoretical robustness guarantee for message-passing GNNs under partially asynchronous (Hogwild-style) inference. We then propose an energy-driven implicit GNN architecture: it formulates message aggregation via an energy function and solves representations through provably convergent fixed-point iteration. This design ensures both theoretical convergence and strong empirical performance. On multi-agent synthetic tasks, our method significantly outperforms existing implicit GNNs; on real-world graph benchmarks, it achieves competitive accuracy. Our work thus bridges theory and practice by providing formal robustness guarantees for asynchronous GNN inference while delivering state-of-the-art performance across diverse evaluation scenarios.

This work addresses the limitation of existing graph neural networks, which rely on fixed message-passing schemes and struggle to adapt to diverse local graph structures. The authors propose an adaptive information flow control framework wherein each node is modeled as an agent governed by a shared, end-to-end trainable policy. Through a dynamic predict–act–observe–correct interaction loop, these agents modulate information propagation in a context-aware manner. The approach introduces a novel tripartite action mechanism—comprising source reception, channel selection, and gain-aware halting—to enable fine-grained control over message passing. Extensive experiments demonstrate consistent and significant improvements over baseline methods across node classification, graph classification, and cross-domain transfer tasks, highlighting the model’s strong generalization capability to complex and varied graph topologies.

Traditional surrogate models struggle to simultaneously optimize both the structure and parameters of supply chains and lack generalization capabilities for graph-structured systems. This work presents the first systematic exploration of graph neural networks (GNNs) as surrogate models for supply chain design, introducing a programmatically generated supply chain graph dataset and leveraging a custom simulation library, SupplyNetPy, to produce large-scale training data for end-to-end differentiable modeling. The proposed approach enables performance prediction at both node and network levels, effectively balancing predictive accuracy and computational cost. By supporting gradient-based topology optimization and sensitivity analysis, this method establishes a foundation for novel design-space exploration strategies, significantly enhancing the efficiency of supply chain configuration and optimization.

This work investigates how graph neural networks (GNNs) can efficiently solve combinatorial optimization problems—such as the Traveling Salesman Problem—under a fully unsupervised, search-free, and non-sequential decision-making setting. The proposed non-autoregressive GNN model incorporates global structural constraints as inductive bias during training, enabling it to generate high-quality solutions in a single forward pass without requiring supervision or explicit search. By leveraging dropout at inference time together with snapshot ensembling, the model substantially narrows the optimality gap and enhances performance through solution diversity. This study presents the first demonstration that GNNs can function as internalized unsupervised heuristics, establishing a novel paradigm for learned combinatorial optimization solvers.

This study systematically investigates the effectiveness boundaries and practical value of Graph Neural Networks (GNNs) across twelve application domains. Building upon a unified design space, it derives both spectral and spatial formulations of GNNs from first principles, analyzes their expressive power through the lens of the Weisfeiler–Leman test, and evaluates domain-specific graph construction strategies and architectural choices. The work establishes the first cross-domain analytical framework that disentangles genuine performance gains from baseline biases, uncovering common challenges such as heterophily, scaling effects, and deployment gaps. It clarifies the applicability limits of GNNs, highlights the discrepancy between leaderboard-topping models and deployable ones, and offers constraint-aware practical guidelines to address issues including oversmoothing, over-squashing, and distributional shifts.