Securing Machine Learning Supply Chains
Abstract
Recent cyber-attacks on supply chains such as the large-scale SolarWinds attack are gaining the attention of cybersecurity experts. Supply chain attacks are growing in frequency and are taking advantage of the trust that organizations put in the dependencies of their supply. The machine learning supply chain is incredibly vulnerable to this category of attack because of the large number of dependencies utilized. We demonstrate a weakness in a machine learning supply chain by attacking the models parameters. We then demonstrate how an organization can implement secure checkpoints that generate integrity metadata and detect this class of attack before proceeding to the next phase in the supply chain.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2021
- Accession Number
- AD1164482
Entities
People
- Joshua D. Strubel
Organizations
- Naval Postgraduate School