Concorde：基于解耦能量最小化的几何感知链接预测框架
Concorde: Geometry-Aware Link Prediction via Decoupled Energy Minimization
AI 导读
arXiv 论文提出 Concorde，一种能量驱动的链接预测框架，将链接似然分解为三个解耦分支：无参数谱锚定、可学习双曲与球面乘积流形上的混合曲率残差，以及双曲空间切丛上的连续向量场。该框架通过学习门控融合并用 N-pair 排序损失训练，在五个覆盖稀疏、稠密、无特征和属性图的基准上取得 SOTA 早期检索精度，稀疏场景下 P@10% 较最强基线接近翻倍。论文被 NeurIPS 2026 接收。
正文 · 原文
Abstract:Link prediction remains difficult in sparse graphs, where diffusion heuristics lack connectivity and geometric methods suffer spectral instability when coupling learned representations with diffusion operators. We introduce Concorde, an energy-based framework that resolves this by decomposing link likelihood into three structurally decoupled branches: (i) a parameter-free spectral anchor that establishes permutation-invariant topological stability; (ii) a mixed-curvature residual on a learnable $\mathbb{H}^{d_h}_{c_h} \times \mathbb{S}^{d_s}_{c_s}$ product manifold, regularized by a feature-aware Ollivier-Ricci curvature proxy; and (iii) a continuous vector field on the tangent bundle of hyperbolic space for non-local structural alignment across topological gaps. The branches are fused via learned gating and trained with an N-pair ranking loss aligned with retrieval-style evaluation. Across five benchmarks spanning sparse, dense, featureless, and attributed regimes, Concorde establishes state-of-the-art early retrieval precision, nearly doubling P@10% over leading baselines in sparse settings.
| Comments: | NeurIPS 2026 |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| ACM classes: | I.2.6; G.2.2 |
| Cite as: | arXiv:2610.36204 [cs.LG] |
| (or arXiv:2610.36204v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.36204 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Ege Demirci [view email]
[v1]
Mon, 26 Oct 2026 19:41:23 UTC (53 KB)
来源：arXiv cs.AI · arxiv.org