Operationalizing Informational Invariance: A Probabilistic Model for Pattern Persistence Under Adversarial Attack

Read full text Summarise with VIRAT
Authors: Fernando Vidal Porto
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This paper presents an operational model for the concept of informational invariance, proposed in previous work (Vidal Porto, 2026a, 2026b). The model quantifies pattern persistence through a probabilistic score I'(x) = (Pcorr - Padv)/(Pcorr + Padv + ε), where Pcorr is the frequency of an element in correlated (preserved) versions and Padv its frequency in adversarial (attacked) versions. Three adversarial scenarios are defined: upper bound (attack on the original), realistic (attack on the correlated version), and baseline (blind attack). From these scenarios, we define Trans-Lesional Regenerative Capacity Cr = lim(Simorig → 0) (Simrec - Simorig)/(1 - Simorig) and Invariant Robustness Rinv = Cr(Realistic) / Cr(Upper Bound). Binary grid simulations show Cr(Realistic) = 68.3%, Cr(Upper Bound) = 51.7%, and Rinv = 1.32. Validation on the MNIST dataset of handwritten digits yields Cr(Realistic) = 6.1%, Cr(Upper Bound) = 0.1%, and Rinv = 75.63, demonstrating that the invariant is substantially more robust in realistic scenarios. The results confirm the hypotheses of previous work and provide a testable foundation for future applications in medical imaging and biological systems.

Fields of Study & Subjects

Adversarial Robustness in Machine Learning Digital Media Forensic Detection Image Retrieval and Classification Techniques

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
1 January 2026
Language
EN