Skip to content

Sub-Poisson forecasting bottleneck

Home / Open Questions / Sub-Poisson forecasting bottleneck

Background: In neural spike-count forecasting, sub-Poisson neurons exhibit regular firing patterns that are consistently harder to predict than super-Poisson neurons, creating a baseline noise floor for evaluation.

Question / Future Work: Further research is needed to determine whether sub-Poisson predictability limitations arise from intrinsic single-cell biophysics or metric artifacts, and to develop dispersion-aware metrics that decouple these factors.

Why It Matters: Identifying the root of the sub-Poisson noise floor is essential for accurately assessing predictive progress in neural forecasting and avoiding metric-induced bias.

Evidence: These neurons therefore impose a hard noise floor on aggregate metrics: ~0.07 regardless of model quality.

Metadata & Links