The Universal Classifier for Graph Learning

2026-09-28

AI Summary

This study addresses the lack of unified feature and structural representations in graph machine learning, which hinders cross-domain generalization and multi-granularity task support. To overcome these limitations, this work proposes an architecture-agnostic universal graph classifier. The method introduces a unified backbone network that accommodates arbitrary feature and class cardinalities, thereby breaking fixed-dimensionality constraints. Specifically, it maps heterogeneous features into a shared space via 3D latent tensors and designs a similarity-based objective function to jointly integrate node-, edge-, and graph-level predictions. Experimental results demonstrate that the proposed model achieves strong zero-shot transfer performance across a full spectrum of tasks—including node classification, regression, and link prediction—on previously unseen graphs.

Abstract

While foundation models have revolutionized natural language processing and computer vision by leveraging universal vocabularies, Graph Machine Learning (GML) remains fractured due to the absence of a unified feature and structural representation across diverse domains. Existing works claiming to be Graph Foundation Models (GFMs) are typically restricted to node-level predictions or require fixed feature dimensions, failing to provide a truly task-agnostic backbone for the full spectrum of graph learning applications. In this paper, we introduce the Universal Classifier (UC), which supports arbitrary feature and class cardinalities, unifying node-, edge-, and graph-level objectives under a single similarity-based classification objective. The UC reformulates all node-, edge-, and graph-level prediction tasks as maximizing similarity in the latent space: by lifting heterogeneous features and labels into 3D latent tensors, the model learns transferable features independent of specific input schemas. This architecture allows a single pre-trained model to generalize to node classification, node regression, and link prediction across unseen graphs with varying feature semantics. Experiments show strong zero-shot transfer performance across node-, link-, and graph-level tasks.

Problem

• Graph Foundation Models
• Graph Machine Learning
• Universal Classification
• Zero-shot Transfer
• Task-agnostic Backbone

Innovation

• Graph Foundation Model
• Universal Classifier
• Zero-shot Transfer
• Similarity-based Classification
• 3D Latent Tensors