---
title: "Standard Graph Neural Networks need curved semantic manifolds"
date: 2026-04-22
canonical: https://patrickaudley.com/#post-graph-neural-networks-need-curved-manifolds
cross-posted-from: https://www.linkedin.com/feed/update/urn:li:activity:7458177770238681088/
source-platform: LinkedIn
tags: [topological-data-analysis, graph-neural-networks, manifold-learning, knowledge-representation]
mentions: [https://patrickaudley.com/#proj-nonlinear-semantic-graphs,
https://patrickaudley.com/#emergent-knowledge-graphs]
author: Patrick Colm Audley
author-url: https://patrickaudley.com/
license: Creative Commons BY-NC-SA CAv2.5
lang: en
---

# Standard Graph Neural Networks need curved semantic manifolds

> Standard GNNs are structurally constrained when mapping complex text attribution; linear aggregation in flat Euclidean
space inevitably forces semantic drift. Looking to connect with researchers in TDA, geometric deep learning, and
spectral graph theory.

Standard Graph Neural Networks are structurally constrained when mapping complex text attribution: linear aggregation in
flat Euclidean space inevitably forces semantic drift. To map high-dimensional knowledge faithfully you have to
transition to curved semantic manifolds, where the geometry itself carries the relational structure.

Across thirty years of building scientific analysis pipelines — genetics, satellite imagery, multi-continent
high-resiliency financial applications — the through-line has been the same: representations must remain
mathematically faithful to their underlying geometry, or they stop being interpretable the moment the data leaves your
dev set.

I've recently open-sourced a framework that discovers emergent knowledge-graph relations in high-order semantic vector
spaces through manifold learning and spectral analysis. Initial proofs, a small teaser, and the Python codebase live at
[paudley/nonlinear-semantic-graphs](https://github.com/paudley/nonlinear-semantic-graphs); the working paper that
motivates the design is in [Publications](#emergent-knowledge-graphs).

I'm looking to connect with researchers and applied scientists specialising in **Topological Data Analysis**,
**Geometric Deep Learning**, and **Knowledge Representation** — especially anyone working on geodesic aggregation or
spectral graph theory — to push these ideas into robust enterprise deployments.
[Comment on the LinkedIn original](https://www.linkedin.com/feed/update/urn:li:activity:7458177770238681088/) or
[drop me a line directly](#contact).

---

*Originally published 2026-04-22 —
[LinkedIn](https://www.linkedin.com/feed/update/urn:li:activity:7458177770238681088/). Canonical version at
<https://patrickaudley.com/#post-graph-neural-networks-need-curved-manifolds>. Author:
[Patrick Colm Audley](https://patrickaudley.com/). *