PAI-MACHINE - Synthesis of machine-optimized cryptographic protocols with applications in secure machine learning systems
Abstract
Cryptography is a fundamental tool in cybersecurity. Without it, we can t have secure communication or authentication between remote phones, vehicles, ships, planes or satellites. With new or looming attacks from high-performance computing or quantum computing, cryptography needs to get more complex as well. Cryptography must adapt, by coming up with more complex constructions and protocols. With complexity comes an increase in running time that reduces the usability and applicability of new secure protocols.What if therewas a tool that could optimise cryptographic protocol implementations to run in the shortest time possible by restructuring its internal code according to the communication, memory and processor limitations in its environment? Such a tool will avoid the manual optimisation pass for each protocol and each hardware target. Security researchers could quickly benchmark new protocols and security engineers can get complex protocols running efficiently in demanding environments like edge devices, vehicles or satellites--independent of the underlying hardware platform or processor.Our roadmap foresees a multi-year research roadmap that will impact the development secure machine learning tools, new protocols using post-quantum cryptography, code generation integrating the abilities of up-and-coming hardware architectures, e.g. ones based on trusted computing or hardware-accelerated primitives.The PAI-MACHINE project is a first step on this path. In the base grant project, we will generalize and scale up the ability to synthesize efficient code forrunning cryptographic protocols from scalar operations to full algorithms. Success on the fundamental research can lead to efficient implementations of cryptographically secure machine learning protocols. The project has a potential for further work, expanding the technique to new research domains and applications.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 24, 2023
- Source ID
- N629092312046
Entities
People
- Dan Bogdanov
Organizations
- Cybernetica
- Office of Naval Research
- United States Navy