AI Computations using physical principles

Speaker: Dr. Kirill Rivkin (RKMAG Corporation, Delaware, USA) 

Abstract: Within the last two decades, we saw a massive introduction of AI algorithms, redefining such fields as computer vision and data processing.When it comes to Physics it created three new venues: devices combining conventional technology with AI processing, DNN accelerated modeling tools, and a demand for massively parallel, low precision and low power computational infrastructure, a natural fit for physics-based analogue computing.In this presentation, we will first demonstrate how a combination of generative AI and optics can result in computer vision systems with exceptional capabilities, followed by a discussion on how the relevant computations can be performed using basic physical principles, with a particular emphasis on previously proposed optics and magnetics-based solutions. Identifying the general failure mechanisms currently preventing a further practical implementation of physics-based computing, we propose how those can be addressed by a specific spin wave-based design, which, for the first time, appears to be capable of solving large-scale practical problems.

Acknowledgments: The scientific visit, during which this lecture was given, is financed by Adam Mickiewicz University Poznan from the funds of the ID-UB program – Call No. 175 “Excellence Initiative Research University Programme”.

Join to the meeting Microsoft Teams:
meetup-join

Google Calendar

iCal Export