NervePool: A simplicial pooling layer built on the nerve complex, published with open code
Pseudomanifold · x · 2026-11-17
Bastian Rieck and collaborators published NervePool in Computing in Geometry and Topology: a pooling layer for simplicial complexes — generalizations of graphs that capture higher-order relationships — with open-source code.
- How it works: cluster the vertices (bring your own clustering), grow each cluster, then take the nerve, yielding a smaller complex that preserves higher-order structure. The topology-motivated set-theoretic construction (unions of stars + nerve complex) is realized as deterministic matrix operations on learned vertex cluster assignments.
- Results: ranks 2nd on graphs behind DiffPool, but shines on simplicial complexes where it can exploit higher-order structure ("they don't know I can see triangles").
- Team: Sarah McGuire Scullen (PNNL), Ernst Röell (Helmholtz Munich/TUM), Liz Munch and Matthew Hirn (Michigan State).
More from Research
- Sharpa's SALA aligns latent actions across human and robot video for dual-axis scaling — Jiachun Jin · 2026-11-17
- OutfitterBench tests LLM agents on 200-day rental business, exposing commitment failures — Jie Huang · 2026-11-17
- Craftax framework evolves code policies via feedback cascade, reaching final stage 50% of time — Víctor Gallego · 2026-11-17
- cuPotential MD benchmark: 3-4x faster than stock OpenMM on an RTX 4090 — gdefabritiis · 2026-11-17
- Four calibration papers reviewed: model confidence can't gate risky agent tool calls — Xjjjjyn · 2026-11-17
- Connectomics nonprofit E11 Bio is hiring across imaging, ML and wet-lab roles — anderssandberg · 2026-11-17