ICLR 2026 Orals

Exchangeability of GNN Representations with Applications to Graph Retrieval

Kartik Nair, Indradyumna Roy, Soumen Chakrabarti, Anirban Dasgupta, Abir De

Graph Learning Thu, Apr 23 · 11:42 AM–11:52 AM · 203 A/B Avg rating: 6.00 (4–8)
Author-provided TL;DR

It shows that graph representations are exchangeable random variables which can help in LSH in graphs

Abstract

In this work, we discover a probabilistic symmetry, called as exchangeability in graph neural networks (GNNs). Specifically, we show that the trained node embedding computed using a large family of graph neural networks, learned under standard optimization tools, are exchangeable random variables. This implies that the probability density of the node embeddings remains invariant with respect to a permutation applied on their dimension axis. This results in identical distribution across the elements of the graph representations. Such a property enables approximation of transportation-based graph similarities by Euclidean similarities between order statistics. Leveraging this reduction, we propose a unified locality-sensitive hashing (LSH) framework that supports diverse relevance measures, including subgraph matching and graph edit distance. Experiments show that our method helps to do LSH more effectively than baselines.

One-sentence summary·Auto-generated by claude-haiku-4-5-20251001(?)

Graph embeddings exhibit exchangeability property, enabling efficient graph retrieval via transport-based similarity approximation with locality-sensitive hashing.

Contributions·Auto-generated by claude-haiku-4-5-20251001(?)
  • Discovery of exchangeability property in trained GNN node embeddings across broad family of architectures
  • Concentration bound for reducing transport problems on node embeddings
  • GRAPHHASH framework for approximate graph retrieval using general transport-based distances
Methods used·Auto-generated by claude-haiku-4-5-20251001(?)
  • exchangeability analysis
  • locality-sensitive hashing
  • order statistics
  • transport distances
Limitations (author-stated)·Auto-generated by claude-haiku-4-5-20251001(?)

Authors did not state explicit limitations.

Future work (author-stated)·Auto-generated by claude-haiku-4-5-20251001(?)
  • Explore other consequences of exchangeability on learning and training dynamics
    from the paper
  • Extend framework to similarities over richer classes of similarity functions for 3D molecular structures
    from the paper

Author keywords

  • GNN
  • Locality sensitive hashing

Related orals

Something off? Let us know →