Data Poisoning Attacks on Informativity for Observability: Invariance-Based Synthesis
Data Poisoning Attacks on Informativity for Observability: Invariance-Based Synthesis
Authors: Iori Takaki, Ahmet Cetinkaya, Hideaki Ishii Date: 2026-04-13 Paper ID: openalex:2604.11657
Summary
This paper investigates data poisoning attacks targeting informativity-based observability analysis in data-driven control. The authors analyze how invertible linear transformations on finite time-series data matrices can manipulate the invariant subspaces that underpin informativity, potentially invalidating system state observability. A constructive synthesis method is provided alongside feasibility conditions, allowing for the generation of minimum-norm perturbations that degrade observability certificates. Numerical experiments confirm that structured, low-magnitude transformations effectively compromise these data-driven security guarantees.
Key Contributions
- Introduces a formal framework for cyber attacks on informativity-based observability analysis in data-driven control systems.
- Demonstrates that invertible linear transformations of data matrices can be used to embed malicious states into the invariant subspace of the dataset.
- Formulates and solves a minimum-norm optimization problem to compute the smallest data distortion necessary to invalidate informativity certificates.
Links
Metadata & Links
- url
- https://arxiv.org/abs/2604.11657
- paper_id
- 2604.11657
- paper_source
- openalex
- domain
- time-series
- tags
-
- architectures
-
- datasets
-
- concept_slugs
-
- dataset_slugs
-
- skill
- TimeSeriesSkill
- created_at
- 2026-04-16T05:47:07Z