Logically Synchronous AI (LSAI): Towards Safe, Secure, and Sustainable AI
Partha Roop
University of Auckland, New Zealand
Abstract
Logical synchrony is a world-first architecture designed by Google DeepMind in cooperation with Auckland and Stanford University. This allows distributed devices to coordinate via logical time synchronisation without the need for a global time master. Logically synchronous compute for AI, can operate at massive, distributed scale, will be fault tolerant, secure, safe and deterministic. In this talk, I will introduce logical synchrony along with a programming language called Timetide and a middleware called Softtide, which enable LSAI to be developed, verified and deployed at competitive scale. LSAI is being developed jointly with a novel temporal logic, called synchronous signal temporal logic (SSTL). SSTL can be used to specify and verify properties of LSAI systems, especially being developed for autonomous vehicle (AV). We will highlight the real-world applicability of LSAI through the running example of an AV, especially through car following and lane changing tasks.
Biography
Partha is the Professor and Head of the Department of Electrical and Computer and Software Engineering at the University of Auckland, New Zealand. He is also the Co-Editor-in-Chief of ACM Books. Partha's research interests are in safety of AI, ethical AI, and AI applications in cyber-physical systems in digital health, autonomous systems and Machine Learning for Real-Time Systems. His work has both academic and industrial uptake, including recent work on Logical Synchrony Networks with Google and cardiac digital twins with MathWorks.