Skip to content

Automated hyperparameter optimization for reservoir computing

Home / Open Questions / Automated hyperparameter optimization for reservoir computing

Background: Reservoir computing models, particularly those designed for memristive hardware, rely on fixed internal weights that are highly sensitive to architectural hyperparameters.

Question / Future Work: The reliance on manual, trial-and-error hyperparameter tuning in gradient-free architectures constitutes a significant barrier to scaling reservoir models across diverse datasets. Research is required to define systematic, computationally efficient tuning procedures that avoid the overhead of traditional derivative-based optimization.

Why It Matters: Without systematic optimization, the performance ceiling of reservoir-based neuromorphic systems remains unpredictable across new problem domains.

Evidence: Furthermore, MARS has plenty of hyperparameters and only a few learnable parameters. The hyperparameters are currently selected manually by try-and-error.

Metadata & Links