Exact Temporal Adversarial Radii for Secure State Estimation with Interval Evidence: Peak, Persistent and Total Corruption Budgets, Coupled Dynamics, and Certified Dynamics Trust
A decision certified from redundant sensor evidence is usually judged one instant at a time. When states are linked by physical dynamics, an adversary must also explain the past. We define, for every decision, four exact temporal adversarial radii (point, peak, persistent and total corruption budgets) and proven-lie values, prove that they are ordered, never decrease with a longer history and are tight, give a linear-time exact algorithm for the peak and total radii and its equivalence with the exact f-robust set-membership filter, prove that the persistent radius is NP-hard even for a scalar state with interval evidence yet fixed-parameter tractable, extend the results to coupled position-velocity dynamics through a Helly-type conflict and a sound projection bound, and turn the dynamics bound into certified numbers that separate model error from attack. The deterministic backbone is machine-checked in Lean 4 (53 theorems), and on 47 real flights of the public ALFA dataset the declared model produced no false alarm. Preprint, version 1, not peer reviewed. AI assistance is disclosed in the paper.
Authors
- Anthony Moreno
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-19
- DOI
- https://doi.org/10.5281/zenodo.24377672
- Primary Topic
- Adversarial Robustness in Machine Learning
- Type
- preprint