Concept icon
Concept

Choosing Between Bias, Variance, and Dataset Mismatch Fixes

A model improvement plan should depend on the dominant source of error. If the main issue is avoidable bias, you would use techniques that make the model fit the training problem better. If the main issue is high variance, you would use methods that improve generalization. Error estimates can also help reveal whether a train/dev mismatch is the real obstacle, so the next step is to target the right kind of change instead of applying generic fixes.

0

1

Concept icon
Updated 2026-08-12

Contributors are:

Who are from:

Tags

Machine Learning

Deep Learning

Supervised Learning

Dive into Deep Learning @ D2L

Data Science

Machine Learning Strategy

Machine Learning Yearning @ DeepLearning.AI

Related