← Home

Daily arXiv Papers

Graph Neural Networks · Graph Learning · LLM × Graph
Conferences
Archives
2026-08-11 2026-08-10 2026-08-09 2026-08-08 2026-08-07 2026-08-06 2026-08-05 2026-08-04 2026-08-03 2026-08-02 2026-08-01 2026-07-31 2026-07-30 2026-07-29 2026-07-28 2026-07-27 2026-07-26 2026-07-25 2026-07-24 2026-07-23 2026-07-22 2026-07-21 2026-07-20 2026-07-19 2026-07-18 2026-07-17 2026-07-16 2026-07-15 2026-07-14 2026-07-13 2026-07-12 2026-07-11 2026-07-10 2026-07-09 2026-07-08 2026-07-07 2026-07-06 2026-07-05 2026-07-04 2026-07-03 2026-07-02 2026-07-01 2026-06-30 2026-06-29 2026-06-28 2026-06-27 2026-06-26 2026-06-25 2026-06-24 2026-06-23 2026-06-22 2026-06-21 2026-06-20 2026-06-19 2026-06-18 2026-06-17 2026-06-16 2026-06-15 2026-06-14 2026-06-13 2026-06-12 2026-06-11 2026-06-10 2026-06-09 2026-06-08 2026-06-07 2026-06-06 2026-06-05 2026-06-04 2026-06-03 2026-06-02 2026-06-01 2026-05-31 2026-05-30 2026-05-29 2026-05-28 2026-05-27 2026-05-26 2026-05-25 2026-05-24 2026-05-23 2026-05-22 2026-05-21 2026-05-20 2026-05-19 2026-05-18 2026-05-17 2026-05-16 2026-05-15 2026-05-14 2026-05-13 2026-05-12 2026-05-11 2026-05-10 2026-05-09 2026-05-08 2026-05-07 2026-05-06 2026-05-05 2026-05-04 2026-05-03 2026-05-02 2026-05-01 2026-04-30 2026-04-29 2026-04-28 2026-04-27 2026-04-26 2026-04-25 2026-04-24 2026-04-23 2026-04-22 2026-04-21 2026-04-20 2026-04-19 2026-04-18 2026-04-17 2026-04-16 2026-04-15 2026-04-14 2026-04-13 2026-04-12 2026-04-11 2026-04-10 2026-04-09 2026-04-08 2026-04-07 2026-04-06 2026-04-05 2026-04-04 2026-04-03 2026-04-02 2026-04-01 2026-03-31 2026-03-30 2026-03-29 2026-03-28 2026-03-27 2026-03-26 2026-03-25 2026-03-24 2026-03-23 2026-03-22 2026-03-21 2026-03-20 2026-03-19 2026-03-18 2026-03-17 2026-03-16 2026-03-15 2026-03-14 2026-03-13 2026-03-12 2026-03-11 2026-03-10 2026-03-09 2026-03-07 2026-03-06 2026-03-05 2026-03-04 2026-03-03 2026-03-02 2026-03-01 2026-02-28 2026-02-27 2026-02-26 2026-02-25 2026-02-24 2026-02-23 2026-02-22 2026-02-21 2026-02-20 2026-02-19 2026-02-18 2026-02-17 2026-02-16 2026-02-15 2026-02-14 2026-02-13 2026-02-12

Showing 23 papers for 2026-07-24

A Graph Neural Network approach to zero-shot Digital Twins
GNN Graph Learning

We propose Zero-Shot Digital Twins that couple real-time visual perception with a geometry-agnostic, physics-informed reasoning engine. The core is a Thermodynamics-Informed Graph Neural Network (TIM-GNN) that operates on flexible geometry, enabling adaptation to changing domains or boundary conditions without retraining.

HypNO: A Graph-Based Neural Operator with Physics-Informed Message Passing for Hyperbolic Conservation Laws
GNN Graph Learning

HypNO is a graph-based neural operator for scalar hyperbolic conservation laws that operates on a space-time graph of finite-volume cells. It uses adjacency-factored, physics-informed message passing to enforce upwinding and entropy conditions near shocks. It is benchmarked on LWR and ARZ traffic-flow models, achieving accurate solution snapshots.

SenCos-GEM: SENet-Calibrated and Law-of-Cosines-Constrained Geometry-Enhanced Molecular Representation for Property Prediction
GNN Graph Learning

SenCos-GEM introduces SENet-calibrated and law-of-cosines-constrained geometry-enhanced molecular representations for property prediction. It addresses noise from coarse empirical force fields in 3D GNN SSL pretraining and lacks explicit physical constraints; the method imposes physics-inspired constraints and leverages dynamic feature modulation during downstream adaptation to improve predictions.

Cardinality-Decomposed Loss: Matching Training Objectives to Relation Structure in Heterogeneous Recommendation Graphs
GNN Graph Learning

We study heterogeneous bipartite graphs in recommendation with varying relation cardinalities. Traditional loss (e.g., BPR) applied uniformly can cause attribute embeddings to collapse. We propose Cardinality-Decomposed Loss that tailors objectives to the specific cardinality of each relation, improving embedding quality and recommendation performance.

Multilevel Graph Wavelet Compressed Sensing with Scale-Aware Neural Recovery
Graph Learning Graph Theory

We propose Graph Wavelet Compressed Sensing (GWCS) for offline graph-signal compression using spectral graph wavelet transform. The framework combines hierarchical graph wavelets with scale-aware neural recovery to enable data-efficient training and accurate recovery of graph signals.

A Polynomial Architecture-Attribution Co-Design Framework for Exact Aumann-Shapley Attribution in GNNs
GNN Graph Learning

We present APEX, a polynomial architecture-attribution co-design framework that makes Aumann-Shapley attribution exactly computable for GNNs under a polynomial formulation. This reduces the computational burden while providing exact, principled attributions for feature and/or node contributions.

Demographically-Informed Heat-Mortality Risk Curves via Risk Graph Neural Networks
GNN Graph Learning

Risk Graph Neural Networks (RGNNs) deliver demographically informed heat-mortality risk curves by using granular census features to tailor the coefficients in a distributed-lag nonlinear model (DLNM). The hierarchical GNN encoder yields interpretable exposure-response curves that reflect population and geographic context.

Graph Learning on Ensembles of Cyclic Peptides: An Investigation of Molecular Ensemble Modeling
GNN Graph Learning

EnsembleEGNN is a molecular ensemble foundation model that encodes an ensemble of conformers by sharing EGNN layers per conformer and pooling with a Set Attention Block. Pretrained on CREMP cyclic peptide ensemble with a multi-task SSL objective to learn from ensembles for improved molecular property prediction.

HERMES: Heterogeneous Edge-Relational Multi-Head Embedded SSM Attention for Traffic Conflict Prediction at Signalized Intersections
GNN Graph Learning

HERMES is a heterogeneous edge-relational multi-head embedded SSM attention network for traffic conflict prediction at signalized intersections. It models temporal scene graphs with heterogeneous edges (vehicles, pedestrians) and uses SSM-informed multi-head attention to reason about interaction risk.

AI-Driven Multi-Hop Relay Selection for Smart Urban NR-V2X Networks via Learning-to-Optimize Graph Neural Networks
GNN Graph Learning

We propose AI-driven multi-hop relay selection for NR-V2X networks using Learning-to-Optimize GNNs. The method handles sparse RSU density, non-line-of-sight, and dynamic topology to select relay links under flow, capacity, and connectivity constraints, improving reliability and latency.

Label-Free Finite-Volume-Residual Training of Attention Graph Neural Networks for Coupled Thermo-Fluid Fields
GNN Graph Learning

Label-Free Finite-Volume-Residual Training of Attention Graph Neural Networks for Coupled Thermo-Fluid Fields. We train neural surrogates by minimizing residuals of governing equations on the mesh, without labeled data. This reduces data and solver costs for 3D thermo-fluid predictions.

Semi-Supervised Text-Attributed Graph Distillation
GNN Graph Learning

Semi-Supervised Text-Attributed Graph Distillation. TAGs combine graph structure with textual semantics; existing methods struggle with scalability, label scarcity, and multimodal interaction. We propose a distillation framework to better fuse text and graph modalities in semi-supervised settings.

AISE-Bench: A Full-Cycle Curated Benchmark for Information Seeking on Academic Knowledge Graphs
Knowledge Graph Graph Learning

AISE-Bench: A Full-Cycle Curated Benchmark for Information Seeking on Academic Knowledge Graphs. We propose a benchmark for tool-augmented LLM information seeking on academic graphs, addressing realistic user intents, multi-step API planning, parameter filling, grounded answers with references, and end-to-end evaluation of process and results.

Optimizing Hypergraph-Based RAG: Toward Better Fact Extraction and Chunk Retrieval
Graph Learning Graph Theory

HyperGraphRAG improves Fact Extraction and Chunk Retrieval by using HyperGraph instead of GraphRAG, and addresses extraction errors via self-consistency prompting; uses Personalized PageRank on the hypergraph to improve chunk retrieval.

Towards Faithful Graph Explanations with Synergistic Edge Effects via Granular Balls
GNN Graph Learning

SeeExplainer provides faithful graph explanations by considering synergistic edge effects via granular balls. It is a parameter-free explainer for GNNs at instance level.

From Static Bibliometrics to Dynamic Knowledge Graphs: An LLM-Powered Framework for Modernizing Science, Technology, and Innovation (STI) Analytics
Knowledge Graph Graph Learning

From Static Bibliometrics to Dynamic Knowledge Graphs: An LLM-Powered Framework for Modernizing STI Analytics. Proposes dynamic knowledge graphs plus LLMs to capture non-linear knowledge ecosystem dynamics, addressing temporal lag and semantic shallowness in STI analytics.

Improved lower bounds for the Shannon capacity of odd cycles
Graph Theory

Improved lower bounds for the Shannon capacity of odd cycles. We construct large independent sets in powers of C7, C11, C13, yielding improved lower bounds for Theta(C7), Theta(C11), Theta(C13) with numerical values.

Knowledge Graph Re-engineering Along the Ontological Continuum (extended version)
Knowledge Graph Graph Learning

Knowledge Graph Re-engineering Along the Ontological Continuum (extended version). Knowledge graphs are central to AI but vary in modelling practices; reengineering is costly to reuse. The paper discusses principled approaches to reengineer KGs to fit new neuro-symbolic AI requirements and how GenAI can automate parts of the process.

Revisiting Degree-Corrected Spectral Clustering: a Condition-Free Spectral Analysis and Extension
Graph Theory

Revisiting Degree-Corrected Spectral Clustering: a Condition-Free Spectral Analysis and Extension. The work provides a condition-free, purely spectral perspective on the performance of degree-corrected spectral clustering, extending its theoretical understanding beyond probabilistic models.

Fast and Efficient Approximate Nearest Neighbor Search for High-Dimensional LLM Embeddings
Graph Learning

Fast and Efficient Approximate Nearest Neighbor Search for High-Dimensional LLM Embeddings. Submitting to SISAP 2026, the work covers kNN graph construction for 1024-dimensional embeddings and MIPS for unnormalized Llama-3.2-8B features, using Equi-Voronoi Polytopes for quantization and targeted reranking to maintain accuracy.

Search-on-Graph: Iterative Informed Navigation for Large Language Model Reasoning on Knowledge Graphs
Knowledge Graph Graph Learning

We propose Search-on-Graph (SoG), an end-to-end framework that iteratively navigates knowledge graphs to support LLM-based reasoning. It addresses the misalignment between reasoning requirements and path-selection modules by coupling path selection more directly with the reasoning objective, reducing erroneous relations and premature pruning. The approach yields more accurate and efficient KG-informed reasoning on large-scale graphs.

Search-on-Graph-R1: Training Large Language Models to Search Knowledge Graphs with Reinforcement Learning
Knowledge Graph Graph Learning

Search-on-Graph-R1 (SoG-R1) trains an 8B model to search knowledge graphs for KGQA using SFT followed by RL. It uses a frontier teacher guided by each question's gold SPARQL query to scaffold the learning process, enabling compact models to perform graph navigation without expensive frontier-scale inference. This yields cost-effective KGQA performance while maintaining competitive accuracy.

TopoGuard: Graph Theory Based Defenses Against Split-Knowledge Attacks on RAG
Graph Theory

TopoGuard introduces graph-theory-based defenses against split-knowledge attacks in Retrieval-Augmented Generation. It demonstrates that these attacks can be structurally invisible to per-document filters, and proposes a defense that analyzes the retrieved-document graph to detect and block malicious cross-document associations. The approach improves robustness by filtering or reweighting retrievals based on graph-structural signals before LM reasoning.