![[2602.13322] Diagnostic Benchmarks for Invariant Learning Dynamics: Empirical Validation of the Eidos Architecture](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.13322] Diagnostic Benchmarks for Invariant Learning Dynamics: Empirical Validation of the Eidos Architecture
Summary
This paper presents the PolyShapes-Ideal (PSI) dataset and diagnostic benchmarks for evaluating topological invariance in machine learning models, specifically validating the Eidos architecture's performance in geometric integrity tasks.
Why It Matters
Understanding invariant learning dynamics is crucial for advancing computer vision models. This research provides empirical validation of the Eidos architecture, demonstrating that geometric integrity can enhance generalization in machine learning, which may lead to more robust AI systems.
Key Takeaways
- The PSI dataset isolates topological invariance from textural correlations.
- Eidos architecture achieved over 99% accuracy on PSI benchmarks.
- The study supports the 'Form-First' hypothesis regarding geometric integrity.
- Zero-shot transfer capabilities were demonstrated across 30 unseen typefaces.
- The findings could influence future research in robust AI model design.
Computer Science > Computer Vision and Pattern Recognition arXiv:2602.13322 (cs) [Submitted on 10 Feb 2026] Title:Diagnostic Benchmarks for Invariant Learning Dynamics: Empirical Validation of the Eidos Architecture Authors:Datorien L. Anderson View a PDF of the paper titled Diagnostic Benchmarks for Invariant Learning Dynamics: Empirical Validation of the Eidos Architecture, by Datorien L. Anderson View PDF HTML (experimental) Abstract:We present the PolyShapes-Ideal (PSI) dataset, a suite of diagnostic benchmarks designed to isolate topological invariance -- the ability to maintain structural identity across affine transformations -- from the textural correlations that dominate standard vision benchmarks. Through three diagnostic probes (polygon classification under noise, zero-shot font transfer from MNIST, and geometric collapse mapping under progressive deformation), we demonstrate that the Eidos architecture achieves >99% accuracy on PSI and 81.67% zero-shot transfer across 30 unseen typefaces without pre-training. These results validate the "Form-First" hypothesis: generalization in structurally constrained architectures is a property of geometric integrity, not statistical scale. Comments: Subjects: Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG) Cite as: arXiv:2602.13322 [cs.CV] (or arXiv:2602.13322v1 [cs.CV] for this version) https://doi.org/10.48550/arXiv.2602.13322 Focus to learn more arXiv-issued DOI via DataCite (pending registra...
