errors

Definition

Deviations between what a system outputs and what is correct, reported as counts or as an error rate — the complement of accuracy. Keep three things apart: errors in this statistical sense, runtime errors in the software sense, and loss, which is the differentiable surrogate optimized during training. In approximation algorithms errors are further split into one-sided and two-sided, which changes what guarantees you can offer.

Explainers using this term

• Probabilistic Data Structures — Counting Without CountingSpace/Time Trade-offs in Hash Coding with Allowable Errors (Bloom