Persistent Memory in Single Node Delay-Coupled Reservoir Computing.

Area of Science:

• Computational Neuroscience
• Dynamical Systems Theory
• Biologically-Inspired Computing

Background:

• Delays are fundamental in biological systems, influencing dynamics from gene regulation to population interactions.
• These delays, arising from signal propagation constraints, provide functional diversity and are key to biological system operations.
• Reservoir Computing (RC) architectures, like the Delay-Coupled Reservoir (DCR), leverage these dynamics for computation.

Purpose of the Study:

• To extend the computational capabilities of the single-node DCR by addressing its inherent fading memory limitation.
• To introduce a novel architecture, the extended DCR, incorporating trained linear feedback.
• To demonstrate the capacity of the extended DCR for complex nonlinear computations requiring non-fading memory.

Main Methods:

• Development of an extended single-node Delay-Coupled Reservoir architecture.
• Integration of task-specific trained linear feedback into the DCR.
• Numerical simulations and case studies to evaluate computational performance on tasks requiring persistent memory.

Main Results:

• The extended DCR with linear feedback successfully performs computations requiring non-fading memory.
• Task-specific feedback significantly enhances the DCR's ability to retain and utilize past information.
• The extended system demonstrates superior performance on complex nonlinear tasks compared to the standard DCR with fading memory.

Conclusions:

• Adding trained linear feedback to the DCR overcomes the fading memory limitation.
• The extended DCR architecture broadens the scope of solvable tasks, enabling complex computations reliant on historical data.
• This work lays the foundation for developing advanced, biologically-inspired computing devices with enhanced memory functionalities.