Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
DY: Fachverband Dynamik und Statistische Physik
DY 6: Machine Learning in Dynamics and Statistical Physics I
DY 6.3: Vortrag
Montag, 9. März 2026, 10:00–10:15, HÜL/S186
Performing inference with physical response: Reservoir computing with active matter substrates — Mario U. Gaimann1 and •Miriam Klopotek1,2 — 1University of Stuttgart, Stuttgart Center for Simulation Science, SimTech Cluster of Excellence EXC 2075, Stuttgart, Germany — 2WIN-Kolleg of the Young Academy, Heidelberg Academy of Sciences and Humanities, Heidelberg, Germany
We explore questions of real-time inference and forecasting of chaotic signals, re-interpreting them in terms of nonequilibrium physical response, by studying a model of information processing with an active matter substrate used in the reservoir computing (RC) paradigm.
The system becomes robustly optimal for computing in a particular dynamical regime due to its intrinsic ability to relax efficiently, which, under driving, unlocks maximal dynamical diversity and susceptibility to chaotic input signals; the mechanisms include self-healing, multi-step dynamical response, and adaptive morphological reorganization [1,2]. Shifting the system’s response away from direct-agent toward collective variables is key, as evidenced by cross-correlative functions in dynamics [2]. These ideas shed light on self-optimizing inference in bio-inspired or material computing that flexibly exploits dynamics across diverse collective scales.
[1] M. U. Gaimann and M. Klopotek, arXiv:2505.05420 (2025).
[2] M. U. Gaimann and M. Klopotek, arXiv:2509.01799 (2025).
Keywords: reservoir computing; active matter; nonequilibrium response; adaptivity; forecasting