Configure ReCA to effectively capture global (non-local) features

Develop and assess configurations of Reservoir Computing with Cellular Automata that can effectively represent and utilize global, non-local features in data, overcoming the locality constraints inherent to short-depth Elementary Cellular Automata reservoirs (e.g., limited information propagation per iteration). Establish design principles or architectural modifications that enable ReCA to handle globally structured time-series or spatial patterns where dynamic time warping and similar methods are typically effective.

Background

The authors argue that short-step ECA reservoirs primarily capture local features due to the limited propagation speed ('speed of light' constraint), whereas many UCR time-series tasks benefit from global alignment methods like dynamic time warping. Their experimental results suggest ReCA is better suited to local feature problems (e.g., bMNIST) and underperforming on tasks requiring global feature modeling, motivating an explicit need to reconfigure ReCA for global features.