Temporal Mapping of Spatial Data
Background: Reservoir computing (RC) architectures require input data to be transformed into temporal pulse sequences to exploit the reservoir’s dynamical state space.
Question / Future Work: There is a lack of systematic or learned approaches for mapping static spatial data into the temporal domains required by memristor-based reservoirs, currently relying on ad-hoc or manually tuned preprocessing transformations. This bottleneck significantly affects how well features are preserved for the readout layer across diverse hardware-constrained reservoir designs.
Why It Matters: The performance and scalability of memristor-based RC in tasks like computer vision are fundamentally limited by how effectively input features are encoded into dynamical state transitions.
Metadata & Links
- created_at
- 2026-04-26T05:52:56Z