Symbol representations and signal dynamics in evolving droplet computers

We investigate several evolutionary computation approaches as a mechanism to “program” networks of excitable chemical droplets. For this kind of systems, we assigned a specific task and concentrated on the characteristics of signals representing symbols. Given a Boolean function as target functionality, 2D networks composed of 10 × 10 droplets were considered in our simulations. Three different set-ups were tested: Evolving network structures with fixed on/off rate coding signals, co-evolution of networks and signals, and network evolution with fixed but pre-evolved signals. Evolutionary computation served in this work not only for designing droplet networks and input signals but also to estimate the quality of a symbol representation: we assume that a signal leading to faster evolution of a successful network for a given task is better suited for the droplet computing infrastructure. Results show that complicated functions like XOR can evolve using only rate coding and simple droplet types, while other functions involving negations like the NAND or the XNOR function evolved slower using rate coding. Furthermore we discovered symbol representations that performed better than the straight forward on/off rate coding signals for the XNOR and AND Boolean functions. We conclude that our approach is suitable for the exploration of signal encoding in networks of excitable droplets.