On when is Reservoir Computing with Cellular Automata Beneficial?

TL;DR

The paper investigates when Reservoir Computing with Cellular Automata (ReCA) provides real benefits versus encoding-driven performance. It implements a minimal, easy-to-reproduce ReCA design and evaluates it on MNIST/bMNIST and the UCR time-series archive, employing SimExp encoding and ablation testing to separate CA contributions from encoding effects. The findings show strong performance gains on bMNIST with several ECA rules, but also reveal that, on UCR, the SimExp encoding largely drives improvements, highlighting the critical need for ablation analyses. The study concludes that ReCA can be feasible and energy-efficient for edge scenarios, but its effectiveness is task-dependent and highly sensitive to encoding, with local feature emphasis favoring CA advantages and global-feature tasks posing a challenge.

Abstract

Reservoir Computing with Cellular Automata (ReCA) is a relatively novel and promising approach. It consists of 3 steps: an encoding scheme to inject the problem into the CA, the CA iterations step itself and a simple classifying step, typically a linear classifier. This paper demonstrates that the ReCA concept is effective even in arguably the simplest implementation of a ReCA system. However, we also report a failed attempt on the UCR Time Series Classification Archive where ReCA seems to work, but only because of the encoding scheme itself, not in any part due to the CA. This highlights the need for ablation testing, i.e., comparing internally with sub-parts of one model, but also raises an open question on what kind of tasks ReCA is best suited for.

Figures (8)

• Figure 1: Example of ECA with rule 90 with TT, starting from a central cell on, executing 7 time-steps.
• Figure 2: RC as a substrate-independent framework
• Figure 3: Basic network Architecture of an ESN.
• Figure 4: Example of MNIST ReCA solution
• Figure 5: Example of time series being expanded into binary, without losing similarity.