Benedikt Vettelschoss (IMEC)
Benedikt’s work focused on novel approaches to open up new universal computation formalisms with reservoir computing. In particular, he re-investigates the finite state machine (FSM) paradigm and a way to embed its switchable behaviour into the traditional reservoir model system: an ESN. The long-term view is to create multi-reservoir systems in which one reservoir provides the switchable behaviour (the state-transition graph of the FSM), while two (or more) other reservoirs translate the input into the desired excitation signals for this module, and transform the input and state into the desired output. Since the last two functionalities could be provided by previous work on multi-reservoir systems performed at UGent Freiberger et al. [2020], Benedikt’s work focused on the missing (and most difficult) part of this system: embedding a state transition graph into an ESN.
The overarching idea is to embed the desired state transition graph of the finite state machine into the reservoir. A randomly initialized dynamical system can be driven to the edge of stability by means of self-organization. For this step, Benedikt proposed a self-organising plasticity rule and constituted a proof-of-concept, using ESNs as a model system. In particular, his approach can successfully drive a system into the vicinity of a multiple Hopf bifurcation point. From this point, some heteroclinic and excitable connections appear in the associated amplitude equations.
Future work may concern the guided self-organization of a dynamically critical state that yields the desired structure of a predefined state-transition graph. In short, the presented mechanism is a first attempt at shedding light on the potential symbolic computational power of dynamical systems. It remains to be revealed how this mechanism integrates into the zoo of fascinating phenomena that constitute complexity.