Concept icon
Concept

Mismatched Validation and Test Splits Can Make Chance Matter More

When a third-party benchmark uses validation and test sets drawn from different distributions, random variation can have a larger effect on reported performance than it would if both sets were drawn from the same distribution.

0

1

Concept icon
Updated 2026-08-12

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Machine Learning Strategy

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Yearning @ DeepLearning.AI

Related