shapesinfinity.
Learn AI · Beginner · 3 min

Parameters vs Hyperparameters

What numbers does the AI learn, and which do YOU choose?

Parameters and Hyperparameters Flow
Training dataHyperparameters you setTraining process runsModel learns parametersTrained model readyParameters fixed now

You pick hyperparameters first. Training updates parameters. After training, parameters stay fixed.

goes inthe AI workscomes out

In simple words

Parameters (the weights) are numbers the AI learns by itself. Hyperparameters are knobs you turn before training starts, like oven temperature.

The real definition

Parameters are weights the model learns from data during training. Hyperparameters are settings you choose before training begins to control how learning happens.

Like… Making a Cake

Hyperparameters are your recipe choices: oven heat, baking time. Parameters are the cake's final texture and taste—they come from baking, you do not choose them directly.

But: You cannot change a baked cake's taste. You can adjust hyperparameters before retraining with new data.

You see it every day

on your phone

Photo recognition

Your phone uses hyperparameter "learning speed" set by engineers. The AI learns parameters (edge weights) from millions of training images.

out in the world

Weather forecasting

Scientists choose hyperparameter "update rounds" first. Then the model learns parameters from past weather data.

at work

Loan decisions

A bank sets hyperparameter "penalty strength" before training. The AI learns parameters linking income and credit score to risk.

Step by step

  1. 1

    Choose hyperparameters

    You pick learning speed, update rounds, and other AI settings.

  2. 2

    Feed in training data

    Provide examples to help the AI learn what is correct.

  3. 3

    AI adjusts parameters

    The model changes its weights (parameters) based on errors.

  4. 4

    Check and repeat

    Test the model. Adjust hyperparameters if results are poor.

Remember

Parameters: numbers the AI learns during trainingHyperparameters: settings you choose before training startsYou control hyperparameters; training creates parametersChange hyperparameters to improve AI performanceParameters locked after training; retraining needed to change

Memory trick

Parameter = (P)roduct of learning. Hyperparameter = (H)uman choice before start. You set H, AI learns P.

Words

parameter puh-RAM-uh-ter
A number the AI learns from training data.
hyperparameter HY-per-RAM-uh-ter
A setting you choose before training the AI.
training TRAIN-ing
The process where AI learns from data examples.
learning speed LER-ning SPEED
How fast the AI updates its parameters each step.
weight WAYT
A parameter that shows importance of one piece of data.
model MAH-dul
The trained AI program ready to make predictions.

Check yourself

quick check

Which one does the AI learn automatically from data?

Hint

Think: which number changes as the AI learns?

quick check

You are building a photo AI. You must decide learning speed before training starts. Is this a parameter or hyperparameter?

Hint

Hyperparameter = your choice before starting.

think

Why might you change a hyperparameter if your AI model gives poor results?

Show a good answer

Hyperparameters control how the AI learns. Different settings help it learn better from the training data. You adjust them to improve accuracy.

Tell a friend

“Parameters are what the AI learns; hyperparameters are what you control before it learns.”

People also ask

Can you change parameters after training?

Parameters are locked after training. To change them, you must train again with new data or new hyperparameters. Hyperparameters you can adjust before each training run.

What happens if you pick bad hyperparameters?

Bad hyperparameters make the AI learn poorly or slowly. The model gives worse results. You test and adjust hyperparameters to find better settings.

How many hyperparameters does an AI have?

Different AI types have different hyperparameters—usually 2 to 20 common ones. Learning speed, update rounds, and model size are typical examples.