Impact of Data Poisoning Attacks on Feasibility and Optimality of Neural Power System Optimizers
Nora Agah, Meiyi Li, Javad Mohammadi · 2025
Read Impact of Data Poisoning Attacks on Feasibility and Optimality of Neural Power System Optimizers by Nora Agah, Meiyi Li, Javad Mohammadi (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-05727.txt.
Description
Nora Agah, Meiyi Li, and Javad Mohammadi ask how badly poisoned training data can hurt machine-learning tools that help run power grids. These tools solve the grid's power flow puzzle much faster than classic solvers, which matters as weather grows extreme and clean energy varies. A poisoning attack quietly alters the training data so the tool learns to make wrong calls, a risk that had not been studied for this use.
The team compares three designs for the tool: one that penalizes bad answers during training, one that repairs answers afterward, and one that maps inputs straight to answers. They score each design on whether its answers stay workable and near-best as attacks grow. The results lay groundwork for hardening grid tools.
Catalog record
| CREATOR | Nora Agah, Meiyi Li, Javad Mohammadi |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 27 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-05727