Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Can TabPFN Compete with GNNs for Node Classification via Graph Tabularization?

Jeongwhan Choi, Woosung Kang, Minseo Kim, Jongwoo Kim, Noseong Park

TL;DR

The paper tackles graph node classification by reframing graphs as tabular data and introducing TabPFN-GN, which converts node attributes, local and global structural features, and topological encodings into per-node tabular inputs for TabPFN inference. By combining node features, structural patterns, Laplacian or random-walk positional encodings, and optional neighborhood smoothing, TabPFN-GN can perform zero-shot node classification without graph-specific training or reliance on language models. Empirical results across 12 benchmarks show TabPFN-GN is competitive with GNNs on homophilous graphs and consistently outperforms them on heterophilous graphs, validating principled feature engineering as a bridge between tabular and graph domains. The work suggests a practical alternative to graph foundation models and highlights avenues for graph-aware synthetic priors to further improve performance, particularly in homophily-heavy settings and graph classification tasks.

Abstract

Foundation models pretrained on large data have demonstrated remarkable zero-shot generalization capabilities across domains. Building on the success of TabPFN for tabular data and its recent extension to time series, we investigate whether graph node classification can be effectively reformulated as a tabular learning problem. We introduce TabPFN-GN, which transforms graph data into tabular features by extracting node attributes, structural properties, positional encodings, and optionally smoothed neighborhood features. This enables TabPFN to perform direct node classification without any graph-specific training or language model dependencies. Our experiments on 12 benchmark datasets reveal that TabPFN-GN achieves competitive performance with GNNs on homophilous graphs and consistently outperforms them on heterophilous graphs. These results demonstrate that principled feature engineering can bridge the gap between tabular and graph domains, providing a practical alternative to task-specific GNN training and LLM-dependent graph foundation models.

Can TabPFN Compete with GNNs for Node Classification via Graph Tabularization?

TL;DR

The paper tackles graph node classification by reframing graphs as tabular data and introducing TabPFN-GN, which converts node attributes, local and global structural features, and topological encodings into per-node tabular inputs for TabPFN inference. By combining node features, structural patterns, Laplacian or random-walk positional encodings, and optional neighborhood smoothing, TabPFN-GN can perform zero-shot node classification without graph-specific training or reliance on language models. Empirical results across 12 benchmarks show TabPFN-GN is competitive with GNNs on homophilous graphs and consistently outperforms them on heterophilous graphs, validating principled feature engineering as a bridge between tabular and graph domains. The work suggests a practical alternative to graph foundation models and highlights avenues for graph-aware synthetic priors to further improve performance, particularly in homophily-heavy settings and graph classification tasks.

Abstract

Foundation models pretrained on large data have demonstrated remarkable zero-shot generalization capabilities across domains. Building on the success of TabPFN for tabular data and its recent extension to time series, we investigate whether graph node classification can be effectively reformulated as a tabular learning problem. We introduce TabPFN-GN, which transforms graph data into tabular features by extracting node attributes, structural properties, positional encodings, and optionally smoothed neighborhood features. This enables TabPFN to perform direct node classification without any graph-specific training or language model dependencies. Our experiments on 12 benchmark datasets reveal that TabPFN-GN achieves competitive performance with GNNs on homophilous graphs and consistently outperforms them on heterophilous graphs. These results demonstrate that principled feature engineering can bridge the gap between tabular and graph domains, providing a practical alternative to task-specific GNN training and LLM-dependent graph foundation models.

Paper Structure

This paper contains 37 sections, 5 equations, 1 figure, 8 tables.

Table of Contents

  1. Introduction
  2. Motivation 1: Limitations of LLM-dependent graph models.
  3. Motivation 2: Success of Tabularization in Time-Series
  4. Preliminaries & Related Work
  5. Prior-data Fitted Network for Tabular Data.
  6. Graph Neural Networks for Node Classification.
  7. Graph Neural Networks for Node Classification.
  8. Graph Foundation Models
  9. Proposed Method
  10. Graph Tabularization for TabPFN
  11. Node Attributes.
  12. Structural Features.
  13. Positional Encodings.
  14. Final Set of Features.
  15. Node Classification with TabPFN
  16. ...and 22 more sections

Figures (1)

  • Figure 1: TabPFN-GN overview. Graph nodes are transformed into tabular features with node attributes, structural properties, and positional encodings, enabling direct inference via TabPFN.