nucleation

LIT-1811: Neural Sheaf Diffusion: A Topological Perspective on Heterophily and Oversmoothing in GNNs

1 min read

Properties5
StatusActive — read in full 2026-10-29 (NOTE-1332) from arXiv v4; worth reading for the cleanest statement of what a sheaf's restriction maps decide about diffusion: which labellings its harmonic space can separate, class by class of maps, with proofs. The separating sheaves are all built from the labels, and the orthogonal construction works at d = 8, which the paper denies.
Version2
Tagsnetwork-science, mathematics, representation-learning, anthology-candidate
Cited as “paper” byNOTE-1332 — Bodnar et al. — Neural sheaf diffusion
Cited as “source” byTHEORY-1139 — A graph network that learns its sheaf restriction maps from node features at each layer reaches the top, mostly within one standard deviation, of the standard small heterophilic node-classification benchmarks and stays within about one point of the best on homophilic ones; on a synthetic balanced bipartite graph it learns the negative transport the theory prescribes, but whether the maps it learns on real graphs have the geometry the theory credits for its success has not been examined, THEORY-1141 — In the time limit, sheaf diffusion on a graph projects node features onto the harmonic space of its sheaf Laplacian, and which labellings that space can linearly separate depends on the class of restriction maps: a weighted graph Laplacian cannot separate the sides of a balanced bipartite graph, signed one-dimensional maps separate any two classes and no one-dimensional sheaf separates three, while diagonal maps with d ≥ C and orthogonal bundles with C ≤ 2d (d = 2, 4 and, against the paper, 8) separate any labelling; every separating sheaf is built from the labels, THEORY-1220 — A graph network that learns its sheaf restriction maps from node features succeeds on heterophilic graphs because it learns the class-aligned transports its separation theory prescribes: sign flips or rotations across classes, identity within them

Graph View

Bodnar, C., Di Giovanni, F., Chamberlain, B. P., Liò, P., & Bronstein, M. M. (2022). Neural sheaf diffusion: A topological perspective on heterophily and oversmoothing in GNNs. NeurIPS 2022.

Why this is here

A5 set it beside LIT-1820: correspondence maps learned from data, general linear and not constrained to a group, which is where QUESTION-282 asks what fixes a correspondence. LIT-1820 cites it (ref. [21]) as gauge structure with no interaction between edge maps. Learned maps are no more meaningful than chosen ones until validated (QUESTION-130). anthology-candidate: an architecture paper.

The reading (NOTE-1332) bears this out and sharpens it: the maps are functions of each edge’s own endpoint features, so nothing ties them around a cycle, and what fixes them is the classification loss. The theory’s separating sheaves are defined from the class labels, so the paper’s “right sheaf” is fixed by the task, not found.

Backlinks