# TopoGeoScore: A Self-Supervised Source-Only Geometric Framework for OOD Checkpoint Selection

Source: [TopoGeoScore: A Self-Supervised Source-Only Geometric Framework for OOD Checkpoint Selection](<https://www.sota2.com/research/paper/6a02b407dbc836b87d82460f>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-05-30T05:00:11.986Z

Authors: Farid Hazratian, Ali Zia, Hien Duy Nguyen

Published: 2026-05-09T10:46:47.000Z

## Abstract

Out-of-distribution (OOD) robustness is difficult to diagnose when target-domain labels are unavailable. We consider a more restrictive source-only variant of unsupervised accuracy estimation: selecting robust checkpoints using only source-domain representations, with no target samples or target labels. We propose \\textbf{TopoGeoScore}, a source-only geometric scorer for label-free OOD checkpoint selection. Given a trained checkpoint, we construct class-conditional mutual $k$-nearest-neighbour graphs from source embeddings and extract three interpretable signals: a torsion-inspired reduced Laplacian log-determinant for global class-manifold complexity, Ollivier--Ricci curvature for local neighbourhood regularity, and higher-order topological summaries for fragmented connectivity, loops, and global--local inconsistency. Instead of fixing their weights by hand, TopoGeoScore learns a non-negative linear score through a self-supervised objective that enforces invariance under approximately geometry-preserving embedding views and separation from structure-breaking views. The score remains interpretable and uses no target-domain samples or labels. Results across CIFAR-based corruption and distribution-shift benchmarks, ImageNet-C, MNLI$\\to$HANS transfer, and OGBN-Arxiv suggest that source representations contain measurable global--local--topological evidence of robustness, supporting practical checkpoint selection before deployment under distribution shift.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Model Selection on CIFAR-10 10.1 (N=21 ckpts)](<https://www.sota2.com/research/sota/model-selection-on-cifar-10-10-1-n-21-ckpts.md>): Model Selection for Out-of-Distribution Accuracy; CIFAR-10 10.1 (N=21 ckpts); Spearman Correlation (ρ): -0.945.

- [Model Selection for Out-of-Distribution Accuracy on CIFAR-10 N=105 ckpt-sev C](<https://www.sota2.com/research/sota/model-selection-for-out-of-distribution-accuracy-on-cifar-10-n-105-ckpt-sev-c.md>): Model Selection for Out-of-Distribution Accuracy; CIFAR-10 N=105 ckpt-sev C; Spearman Correlation (ρ): -0.641.

- [Unsupervised Model Selection on CIFAR 10.1 (OOD)](<https://www.sota2.com/research/sota/unsupervised-model-selection-on-cifar-10-1-ood.md>): Unsupervised Model Selection; CIFAR 10.1 (OOD); Spearman Rho: -0.945.

- [Unsupervised Model Selection on CIFAR-10-C](<https://www.sota2.com/research/sota/unsupervised-model-selection-on-cifar-10-c.md>): Unsupervised Model Selection; CIFAR-10-C; Spearman Rho: -0.641.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2605.08870>)

[PDF](<https://arxiv.org/pdf/2605.08870.pdf>)
