Securing machine learning models
- Sandia National Laboratories (SNL), Albuquerque, NM, and Livermore, CA (United States)
We discuss the challenges and approaches to securing numeric computation against adversaries who may want to discover hidden parameters or values used by the algorithm. We discuss techniques that are both cryptographic and non-cryptographic in nature. Cryptographic solutions are either not yet algorithmically feasible or currently require more computational resources than are reasonable to have in a deployed setting. Non-cryptographic solutions may be computationally faster, but these cannot stop a determined adversary. For one such non-cryptographic solution, mixed Boolean arithmetic, we suggest a number of improvements that may protect the obfuscated calculation against current automated deobfuscation methods.
- Research Organization:
- Sandia National Laboratories (SNL-NM), Albuquerque, NM (United States)
- Sponsoring Organization:
- USDOE Laboratory Directed Research and Development (LDRD) Program
- DOE Contract Number:
- AC04-94AL85000; NA0003525
- OSTI ID:
- 1661020
- Report Number(s):
- SAND--2020-9254; 690466
- Country of Publication:
- United States
- Language:
- English
Similar Records
A flexible framework for secure and efficient program obfuscation.
Securing mobile code.