Physics-Inspired SAT-Solving through Memcomputing
Chesson Sipling (UCSD)
Monday, January 13th 2025, 2-3pm
Abstract:
Memcomputing machines [1] are novel devices that fall firmly outside the traditional Turing paradigm. Rather than constituting an algorithm, they are carefully designed dynamical systems with memory that naturally evolve to the solution of a given combinatorial problem. To achieve this, the problem (expressed as a Boolean or algebraic circuit) is first mapped into an electronic circuit. Additional “memory” variables are then coupled to the primary variables to induce long-range correlations, enabling the system to evolve collectively. The memory variables also “open up” directions in phase space by transforming all local minima into saddle points. Together, this enables the system to find its equilibrium efficiently with respect to system size. In particular, I will discuss how memory variables can induce long-range correlations in any physical system, explain how Memcomputing circuits are designed, provide topological and numerical arguments for their scalability, and analyze the simulation of these circuits on traditional hardware. Work supported by NSF.
[1] M. Di Ventra, MemComputing: Fundamentals and Applications (Oxford University Press, 2022).
Bio:
Chesson Sipling is a graduate student in the Physics Department at UCSD, advised by Prof. Max Di Ventra. His studies of alternative, physics-based approaches to computation, time non-locality in dynamical systems, and cortical models of the brain have been featured at APS 2024 and published in Nature Communications. Additionally, he is one of the newest recipients of the ARCS Fellowship, which provides unrestricted funding to top STEM graduate students nationwide.