Understanding the Limits of Artificial Intelligence for Warfighters Volume 2, Distributional Shift in Cybersecurity Datasets
Abstract
The occurrence of distributional shift can reduce the performance of machine-learning (ML) systems. This issue is of particular relevance to cybersecurity datasets because the signatures of cyberattacks can shift rapidly and unpredictably in many different ways the data environment is both high-dimensional and highly nonstationary. In seeking a solution to the detection of such a shift along with mitigation, we can create andenhance ML models so that they are more robust and effective. Therefore, detecting and mitigating the adverse effects of distributional shift is paramount to effectively defending against cyberattacks.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 2024
- Accession Number
- AD1218234
Entities
People
- Anthony Jacques
- Erik Van Hegewald
- Gavin S. Hartnett
- Joshua Steier
- Lance Menthe
Organizations
- RAND Corporation