ATLAS: Adaptive Topology -based Learning at Scale for Homophilic and Heterophilic Graphs

ICLR 2026 Conference Submission13693 Authors

Keywords: Topology aware embedding, Scalability, Heterophilic graphs, MLP, GNN

TL;DR: We present a topology aware MLP-based graph learning algorithm as a high-fidelity and scalable alternative to GNNs

Abstract:

We present ATLAS (Adaptive Topology - based Learning at Scale for Homophilic and Heterophilic Graphs), a novel graph learning algorithm that addresses two important challenges in graph neural networks (GNNs). First, the accuracy of GNNs degrades when the graph is heterophilic. Second, the iterative feature aggregation limits the scalability of GNNs to large graphs. We address these challenges by extracting topological information about the graph communities at different levels of refinement, concatenating the community assignments to the feature vector, and applying multilayer perceptrons (MLPs) on this new feature vector. By doing so, we inherently obtain the topological data about the nodes and their neighbors without invoking aggregation. Because MLPs are typically more scalable than GNNs, our approach applies to large graphs—without the need for sampling. Our results, on a wide set of graphs, show that ATLAS has comparable accuracy to baseline methods, with accuracy being as high as 20 percentage points over GCN for heterophilic graphs with negative structural bias and
11 percentage points over MLP for homophilic graphs. Furthermore, we show how multi-resolution community features systematically modulate performance in both homophilic and heterophilic settings, opening a principled path toward explainable graph learning.

Primary Area: learning on graphs and other geometries & topologies

Submission Number: 13693