Berlin 2024 – scientific programme

Parts | Days | Selection | Search | Updates | Downloads | Help

DY: Fachverband Dynamik und Statistische Physik

DY 34: Poster: Machine Learning, Data Science, and Reservoir Computing

DY 34.1: Poster

Wednesday, March 20, 2024, 15:00–18:00, Poster C

A First Approach to Dynamically Solving Quadratic Unconstrained Optimization Problems with Memristive Oscillator Networks — •Bakr Al Beattie and Karlheinz Ochs — Chair of Digital Communication Systems, Ruhr University Bochum, 44801 Bochum, Germany

In recent years, a new computational paradigm based on a network of resistively coupled oscillators has emerged. These devices are referred to as oscillator-based optimizers. They are built, so they have the natural tendency of minimizing an energy function to which quadratic unconstrained binary optimization problems (QUBOs) can be mapped. A challenge of oscillator-based optimization is that the structure of the oscillator network must be changed every time a new QUBO is mapped. This is because the connectivity of the network encodes the coefficients of the optimization problem. To deal with this issue, we propose making use of memristors (memory resistors), which can switch between multiple resistance states. To utilize these devices, it is usually required to have a dedicated programming circuit to set the desired resistance state. In this work, we aim to show that we can omit on using such programming circuits by working with suitable oscillators. To demonstrate this approach, we iteratively solve multiple optimization problems, where we alternate between a programming phase and a solution phase.

Keywords: synchronization; oscillator network; ising machine; optimization; memristor