Acceleration of AI for Implementation Security Testing
OPTIMIST (Open Tools, Interfaces, and Metrics for Implementation Security Testing) is an open-source initiative to improve standardization and interoperability in security testing for cryptographic implementations. AI acceleration platforms and AI algorithms play a role of growing importance in implementation security testing. This session will bring together researchers, industry professionals, and government stakeholders to discuss challenges, methodologies, and best practices in acceleration of AI for Implementation Security Testing.
Talks
Debdeep Mukhopadhyay, Professor, Indian Institute of Technology Kharagpur — Side Channel and Fault Attack Testing of Cryptosystems in the view of Dr AI
Abstract
Machine learning techniques offer promising tools for security evaluations of crypto-implementations. In this talk, we present two developments in this context: first, the evolution of machine learning architectures for performing efficient side channel attacks. Secondly, we discuss the relatively less explored field on developing machine learning based test methodologies for leakage assessment under fault attacks. Finally, we conclude with some general perspectives in making ML-accelerated testing of implementation based security of cryptosystems more effective.
Stjepan Picek, Associate Professor, Radboud University — Machine Learning-based Side-channel Analysis and Evaluation
Abstract
Machine learning (and more recently, deep learning) showed significant potential for side-channel analysis (SCA). Indeed, such techniques can break protected targets while requiring minimal expert knowledge. However, different setups (targets, ciphers, countermeasures, acquisition setups, etc.) may easily necessitate different machine learning architectures or settings. In this talk, we will briefly discuss (potential) advantages of standard architectures and libraries and how those could further push state-of-the-art machine learning-based SCA.
Jakub Breier, Senior Cyber Security Manager, TTControl GmbH — AI-accelerated Implementation Testing: Research vs Practice
Abstract
Leakage assessment plays an important role in evaluating cryptographic implementations in practice. Methods such as test vector leakage assessment (TVLA) have been utilized in the certification laboratories to check if side-channel leakage is present. This talk will outline the current efforts in deep learning-based leakage assessment research and discuss the challenges of using the developed methods in practice.
Working Group Meetings
- 1 May, 3PM GMT (11AM EDT) - Working Group Meeting 1
- 15 May, 3PM GMT (11AM EDT) - Working Group Meeting 2
- 29 May, 3PM GMT (11AM EDT) - Working Group Meeting 3