Optimal balance of nonlinearity and linearity in reservoir responses

Determine the optimal balance between nonlinear and linear response characteristics in reservoir computing substrates so as to maximize time-series forecasting performance and stability across tasks and conditions.

Background

The paper situates active matter reservoir computing within broader unresolved issues in reservoir computing. One foundational question concerns how much nonlinearity versus linearity in the reservoir’s internal dynamics yields the best predictive performance. This trade-off impacts fading memory, separability, and robustness of reservoirs, and has implications for both digital and physical substrates.