Showing 15 papers for 2026-07-17
We propose NeuroGRIP, a retrieval-augmented graph refinement framework for knowledge-grounded EEG seizure diagnosis. It integrates external medical knowledge retrieval with graph refinement to produce more clinically plausible and interpretable graphs. This helps mitigate spurious inter-channel connections common in purely data-driven STGNNs.
LATTICE is a graph-based self-supervised framework for multimodal spatial omics integration. It learns spot-level representations by harmonizing five aligned modality blocks per Visium spot, including Visium RNA, scMultiome RNA, scMultiome ATAC, spatial ATAC, and spatial CUT&Tag.
We propose Grad2Fair, a gradient-driven approach to graph fairness that does not require access to true demographic attributes. It uses predicted demographics as proxies to enforce fairness constraints and analyzes how estimation errors affect fairness.
We introduce GAttNHP, a Group Attention Neural Hawkes Process for extrapolation in temporal knowledge graphs. The model uses group-aware attention within a Hawkes process to capture long-range temporal dependencies, cross-chain interactions, and heavy-tailed inter-arrival times, improving future event forecasting.
ChronoQG presents a temporally expressive and hop-bounded benchmark for temporal KG question generation. It ensures generated questions respect temporal validity, event ordering, and temporal constraints, enabling evaluation of temporal fidelity in KGQG.
We study enhancing small language models reasoning through a neuro-symbolic agentic framework that grounds reasoning in a knowledge graph. Using CLUTRR kinship benchmarks, we adapt Gemma 3 and Llama 3.2 to operate as minimalist agents with two specialized components for reasoning and explanation.
SAGA studies schema-aware grounding for agentic Text-to-SPARQL generation in complex KBQA. Interactive agents alternately reason, query the KB, and extend a partial SPARQL query; the grounding quality depends on what the interaction tools expose. SAGA proposes strategies to improve KB grounding through schema-aware exposures.
What Do Temporal Graph Learning Models Learn? analyzes reliability of benchmarks and what signals models actually use; shows issues with evaluation protocols and surprising competitiveness of simple heuristics. The paper provides analysis to understand the true features driving predictions.
A short review on the maximum clique problem algorithms with classical, AI, and quantum methods. It surveys classical algorithms and recent AI and quantum approaches for finding maximum cliques.
DualHNIE presents dual-channel hypergraph learning for node importance estimation in heterogeneous knowledge graphs. It captures higher-order interactions induced by meta-relational structures and disentangles topology and semantic attributes to improve node importance representations.
Towards Predicting Multi-Vulnerability Attack Chains in Software Supply Chains from SBOM Graphs. It treats SBOM structure and scanner outputs as a dependency graph to learn multi-vulnerability attack chains, enabling proactive security analysis beyond per-CVE studies.
Improving Molecular Property Prediction in Small Language Models Using Graph-based Tools. It proposes a Context-Augmented Prompting framework where a trained GNN expert provides predictive hints and extracts instance-specific explanatory subgraphs, enabling tool use at inference to improve SMILES-based property prediction.
CoDiffGRN: Rethinking Gene Regulatory Network Inference via the BEELINE-KGC Benchmark and Co-evolutionary Discrete Diffusion. The work rethinks GRN inference with a new benchmark and a co-evolutionary discrete diffusion method to improve inductive generalization to unseen genes.
MxGPS: Multiplex Graph Transformers for a Power Grid Foundation Model. It proposes multiplex graph transformers to serve as a power-grid foundation model, addressing topology overfitting and improving robustness under topology shifts.
NodeImport addresses imbalanced node classification by assessing node importance to guide training, improving performance on minority classes.