Design principles for optoelectronic light-scattering reservoir computing at the edge of chaos
Design principles for optoelectronic light-scattering reservoir computing at the edge of chaos
Authors: G KIM, YongKeun Park Date: 2026-05-26 Paper ID: openalex:2605.26848
Summary
This paper establishes systematic design principles for reconfigurable optoelectronic light-scattering reservoir computers by mapping three physical control axes: reservoir dynamics, input coupling, and interconnectivity. The authors identify that memory capacity is maximized at the edge of chaos, evidenced by a near-zero maximal Lyapunov exponent, providing a transferable framework for photonic hardware design. Experimental validation demonstrates that operating at this three-axis optimum enables high-performance chaotic time-series prediction and speech classification.
Key Contributions
- Established a systematic three-axis design framework for optimizing physical light-scattering reservoirs by linking hardware control variables to computational performance.
- Demonstrated that the peak memory capacity in photonic reservoirs coincides with the edge-of-chaos, characterized by a near-zero maximal Lyapunov exponent.
- Achieved 84.5% classification accuracy on the 10-class Speech Commands benchmark using a physical optoelectronic reservoir operated at the identified hardware optimum.
Key Concepts
- optoelectronic-light-scattering-reservoir-computing: A design framework for optimizing physical light-scattering reservoirs by mapping reservoir dynamics, input coupling, and interconnectivity to computational performance.
Archivist Review
I have approved the core conceptual framework for optoelectronic reservoir computing design. I rejected the Speech Commands dataset as it is a well-established benchmark not specific to this research. The review focused on identifying reusable design principles that transcend specific hardware implementations.
Approved Concepts
- Optoelectronic Light-Scattering Reservoir Computing: The paper formalizes the design space for physical optical reservoirs, shifting the field from ad-hoc tuning to systematic optimization.
Rejected Candidates
- [dataset] Speech Commands (
speech-commands) - not_novel: This is a standard, widely used benchmark dataset that does not qualify as a unique or critical contribution needing a permanent vault note.
Links
Metadata & Links
- url
- https://arxiv.org/abs/2605.26848
- paper_id
- 2605.26848
- paper_source
- openalex
- domain
- nlp
- tags
-
- architectures
-
- datasets
-
- concept_slugs
- optoelectronic-light-scattering-reservoir-computing
- dataset_slugs
-
- skill
- TimeSeriesSkill
- created_at
- 2026-05-29T06:29:43Z