Regensburg 2025 – scientific programme

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

DY: Fachverband Dynamik und Statistische Physik

DY 31: Focus Session: Nonequilibrium Collective Behavior in Open Classical and Quantum Systems

DY 31.8: Talk

Thursday, March 20, 2025, 11:45–12:00, H37

Optimal dynamical regimes for reservoir computing with soft matter — •Mario U. Gaimann and Miriam Klopotek — Stuttgart Center for Simulation Science (SimTech), Cluster of Excellence EXC 2075, University of Stuttgart, Germany

Reservoir computing with physical systems is a promising approach for next-generation and in materio computing. Recently, active matter systems for reservoir computing were introduced by Lymburn et al. (Chaos 31(3), 033121, 2021). However, the optimal properties of active matter systems for reservoir computing remain poorly understood. Here we show that viscoelastic, overdamped dynamics yield high predictive performances. This is remarkable since it was previously believed that optimal swarm dynamics are found at a gas-to-liquid phase transition. We relate predictive performance to correlations of agent velocities and their fluctuations. The optimal overdamped swarms show rich phenomenology: interface formation and breaking, local shear thinning, and self-healing. We show that the overdamped regime is optimal across a range of different chaotic attractors. Notably the optimal dynamics are already uncovered by studying reservoir computing with a single particle. Our results demonstrate the importance of tuning basic dynamical properties in physical reservoir substrates to generate optimal correlative effects. Reservoir computing with viscoelastic soft matter inspires novel mechanisms for computing in matter and novel computing devices based on these principles.

Keywords: Reservoir Computing; Active Matter; Non-equilibrium Dynamics; Time-Series Prediction; Machine Learning