Topological deep learning tested on subtle shape properties beyond common features

Beyond Homology: Non-Abelian and Orientability Invariants Sensitivity in Cellular Networks

Computational Geometry

Summary

Topological deep learning models often check if they can recognize certain shape features, but it’s unclear if they truly understand these features or just pick up on accidental details. The authors created pairs of shapes that look the same by usual checks but differ in deeper, more complicated properties. They found that some models can distinguish these tricky differences, but only under specific conditions, and deeper investigation is needed to fully understand what the models learn. This means claims about these models’ abilities should be careful and precise about what they actually measure.

What this means in practice

  • For machine learning engineers: Design graph neural networks that better detect complex topological differences beyond standard homology features in data structures.
  • For network analysis teams: Assess the limits of topological features used in network models by testing if subtle shape differences can be recognized beyond common invariants.

Authors

Raghuwansh Raj, George Panagopoulos, Jun Pang

Abstract

Topological deep learning (TDL) architectures are typically evaluated on synthetic complex pairs that differ in a target invariant. Yet, whether a model recovers that invariant or an artifact of the chosen discretization is seldom audited. To examine this, we design and test pairs of complexes that look identical to standard homological and local message-passing tests, have the same Hasse graphs, yet differ in the target invariant. We study two beyond-homology regimes: nonAbelian fundamental-group structure and homotopy-matched orientability. For the former, we construct genus-2 pairs that match homology, cell statistics, local incidence profiles, and rank-colored Hasse 1-WL behavior while differing in their fundamental groups. For the latter, we evaluate punctured torus-Klein bottle pairs and show that distance, boundary, cycle, and spectral signals can confound apparent orientability detection, motivating stronger matched controls. Under the tested settings, subcomplex-lifted architectures separate some pairs beyond the matched homological, local-incidence, and Hasse 1-WL controls, while a depth sweep reveals an empirical depth-dependent trend without establishing a general barrier or direct computation of the target invariants. These results show that claims of topological expressivity must be stated relative to the model representation, explicitly controlled information channels, and the precise scope of the benchmark.