Working definition
Physical reservoir computing drives a dynamical substrate with time-varying input, samples its high-dimensional transient response, and trains a comparatively simple readout. The substrate may be photonic, mechanical, magnetic, electronic, chemical, or biological. Evaluation requires stability, memory capacity, controllability, readout cost, reset behavior, and fair digital baselines.
Mechanism
- Inject a time-varying input into a nonlinear substrate.
- Measure an expanded transient state.
- Train and validate a readout on held-out sequences.
Measurements
- Task loss
- Memory capacity
- Drift, reset time, and readout cost
Reproducibility controls
- Version hardware, software, firmware, and analysis code.
- Declare dataset, preprocessing, random seeds, and measurement boundary.
- Report repeated runs, variation, exclusions, and failed trials.
Limits and failure modes
- Physical complexity is not useful computation by itself.
- Readout and instrumentation may dominate.
Mathematical connection
Formal structure without substrate erasure
Transient substrate response
Characterize nonlinear response, fading memory, stability, and separability for a declared input regime.
Inputs
- Input sequence
- Measured substrate state
- Readout protocol
Outputs
- State-space features
- Memory estimate
- Stability region
Limit: Rich dynamics are not useful computation until a held-out task and complete readout cost demonstrate value.
Technical and governance sources
- [1]NeuroBench: Advancing Neuromorphic Computing Through Collaborative, Fair and Representative Benchmarking · National Institute of Standards and Technology
Establishes: A community framework separating algorithm and system tracks and defining task, correctness, efficiency, and reporting procedures intended to make neuromorphic results more comparable and reproducible.
Boundary: A benchmark ranks submitted systems on declared tasks and metrics. It does not prove general intelligence, biological equivalence, safety, usefulness outside the benchmark, or superiority under unreported host and data costs.