Universality of Simple Cycle Reservoirs
Date:
This talk presents the material of two papers: “Simple Cycle Reservoirs are Universal” and “Universality of Real Minimal Complexity Reservoir”.
The abstract of the paper is included for completeness:
Reservoir computation (RC) models form a subclass of recurrent neural networks with fixed non-trainable input and dynamic coupling weights. Only the static readout from the state space (reservoir) is trainable, thus circumventing the issues associated with backpropagating error signals through time. Simple Cycle Reservoirs (SCR) represent a specialized class of RC models with a highly constrained reservoir architecture with only one degree of freedom, characterized by uniform ring connectivity and binary input-to-reservoir weights with an aperiodic sign pattern. These architectures are particularly well-suited for implementation on photonic integrated circuits, enabling high-performance, low-latency processing in various practical tasks. This talk presents the results of two recent papers demonstrating that SCRs are universal approximators of time-invariant dynamic filters with fading memory in and , respectively. The universality of SCRs not only not only advances theoretical understanding but are also pivotal for practical hardware implementations in emerging computing technologies.
This is a joint work with Prof. Boyu Li (New Mexico State University, USA) and Prof. Peter Tiňo (University of Birmingham, UK).
The slides are available at: