Operational Feature Fingerprints of Graph Datasets via a White-Box Signal-Subspace Probe

Abstract:

Graph neural networks achieve strong node-classification performance, but learned message
passing entangles ego features, neighborhood smoothing, high-pass graph differences, class
geometry, and classifier-boundary effects inside opaque representations. This makes it difficult
to determine why nodes are classified as they are, and which graph-learning mechanisms
are useful, harmful, or necessary for a given dataset. We propose WG-SRC (White-box
Graph Signal–Subspace Residual Classifier), a white-box signal-subspace probe for prediction
and graph dataset diagnosis. WG-SRC replaces learned message passing with an explicit,
named graph-signal dictionary containing raw features, row- and symmetric-normalized
low-pass propagation, and high-pass graph differences. It then combines Fisher coordinate
selection, class-wise PCA subspaces, closed-form multi-α ridge classification, and validation-
based score fusion. Because every signal block and decision module is explicit, the fitted
scaffold produces both predictions and an operational fingerprint over raw-feature, low-pass,
high-pass, class-geometric, and ridge-boundary mechanisms. Across six node-classification
datasets, WG-SRC remains competitive with aligned reproduced baselines and achieves a
positive average gain under matched repeated splits. Its fingerprints distinguish low-pass-
dominated Amazon graphs, mixed high-pass and class-geometrically complex Chameleon
behavior, and raw- or boundary-sensitive WebKB graphs. Aligned interventions further
show that these fingerprints are operational: they identify when high-pass blocks behave
like removable noise, when graph-derived or raw signals should be preserved, and when
ridge-type boundary correction matters. Additional fixed black-box component probes further
show that measured dataset fingerprints organize architectural behavior across multiple
black-box families: different measured dataset conditions repeatedly favor different inductive
biases. Thus, WG-SRC serves both as a functioning white-box classifier and as a dataset-
fingerprinting probe, enabling fingerprint-conditioned analysis of how black-box graph-model
components behave under different measured dataset conditions.

Submission Type: Regular submission (no more than 12 pages of main content)

Assigned Action Editor: ~Devendra_Singh_Dhami1

Submission Number: 8665