Predicting IPC after noise-reduction by ensemble averaging

Determine how to predict, from the original noisy data, the information processing capacity that will result after applying noise-reduction strategies such as averaging repeated state traces.

Background

The paper studies information processing capacity (IPC) in input-driven dynamical systems whose states are corrupted by process, input, or measurement noise. In experimental settings, repeated runs under the same input are commonly averaged to improve the signal-to-noise ratio before evaluating the IPC.

The authors identify as unresolved the problem of predicting the IPC obtained after such noise-reduction procedures directly from the initially observed noisy data. Their CROP method addresses reconstruction of the noise-free component from noisy observations, but the cited sentence frames prediction of the post-averaging IPC as an open problem.