Evaluation · Flagship experience

Bias, Variance & Learning Curves

Is poor validation performance caused by underfitting or instability?

Start here

Is poor validation performance caused by underfitting or instability?

Bias describes systematic simplification; variance describes sensitivity to training data. Learning curves compare training and validation performance as data grows.

Building interactive view…

Understand

Build the mental model

Bias describes systematic simplification; variance describes sensitivity to training data. Learning curves compare training and validation performance as data grows. Diagnose before tuning. More complexity does not solve high variance; more data does not always solve high bias.

Click a stage to inspect what happens, what changes, and what should be checked before moving on.

Stage 1

Model complexity

This is a learning/estimation stage. Separate the data supplied to the algorithm from the parameters or structure it learns, and keep validation information outside the fit. Technical context for Bias, Variance & Learning Curves: High bias often yields similarly poor train/validation scores; high variance yields a large gap. Regularisation, data size and model complexity move the trade-off.

Practitioner checkpoint: Diagnose before tuning. More complexity does not solve high variance; more data does not always solve high bias.

What happens if…?

Break the assumption deliberately

Increase model complexity and watch training error fall before validation error eventually rises.

Experiment intensity45%

Move the control and explain what you expect before reading the visual.

Technical lens

Formalise what the visual is doing

High bias often yields similarly poor train/validation scores; high variance yields a large gap. Regularisation, data size and model complexity move the trade-off.

Technical questionUse a tiny case to make the mechanism observable. High bias often yields similarly poor train/validation scores; high variance yields a large gap. Regularisation, data size and model complexity move the trade-off. Verify one intermediate quantity, state change or mapping independently; then predict the consequence of this change: Increase model complexity and watch training error fall before validation error eventually rises.

Practitioner lens

Use it responsibly

Diagnose before tuning. More complexity does not solve high variance; more data does not always solve high bias.

Transfer testTransfer this idea to a new example and justify each decision using this practitioner rule: Diagnose before tuning. More complexity does not solve high variance; more data does not always solve high bias. Then explain what should change if you deliberately test: Increase model complexity and watch training error fall before validation error eventually rises.

Worked exploration

Use the visual as an experiment, not decoration

Compare a straight line, moderate curve and highly wiggly model on the same training data. Track training and validation error as complexity increases: underfit → useful region → overfit.

Technical lens

High bias often yields similarly poor train/validation scores; high variance yields a large gap. Regularisation, data size and model complexity move the trade-off.

Practitioner check

Diagnose before tuning. More complexity does not solve high variance; more data does not always solve high bias.

Prediction before interaction

Increase model complexity and watch training error fall before validation error eventually rises.

Exploration walkthrough

Turn the interaction into an evidence trail

Compare a straight line, moderate curve and highly wiggly model on the same training data. Track training and validation error as complexity increases: underfit → useful region → overfit. Before moving the control, state your prediction. After the visual changes, name the specific state, statistic, boundary or mapping that changed and explain why that change is consistent—or inconsistent—with your prediction.

Record one observable quantity before the interaction and the same quantity afterwards.
Change one factor at a time so the causal effect of the control is inspectable.
Use an edge or failure case to discover where the concept stops behaving as the simple story suggests.

Reference depth

Open the complete material

The flagship experience is the map. These pages contain the roads.

Deep Learning Hub lessons

Generalisation, bias and varianceMathematical & Statistical Foundations High biasLearning Curves, Bias & Variance High varianceLearning Curves, Bias & Variance Learning curvesLearning Curves, Bias & Variance Complexity curvesLearning Curves, Bias & Variance

Related models & simulations

Use the Concept Atlas for related methods and adjacent concepts.

Continue this exact concept

Choose depth, practice or application.

These destinations are explicitly mapped to Bias, Variance & Learning Curves; they are not generic landing-page fallbacks.