Definition
These are the core words you'll meet in every AI topic. An algorithm is a step-by-step method. A model is what training produces. Data is the examples, which have features (inputs) and labels (answers). Parameters are what the model learns, training is the learning, and inference is using the result.
In simple words
Think of cooking. The data is your ingredients. The algorithm is the recipe. The training is the cooking. The model is the finished dish. Inference is serving it to a guest.
Why it matters
Half of the confusion in AI comes from these words being used loosely. With them clear, the rest of this course reads easily.
How it works
| Word | Meaning | House price example |
|---|---|---|
| Dataset | A collection of examples | 10,000 past house sales |
| Sample / example | One row of data | One house |
| Feature | An input the model looks at | Area, bedrooms, location |
| Label | The right answer to predict | The sale price |
| Algorithm | The learning method | Linear regression |
| Model | The result of training | price = 5000 × area + 2,00,000 |
| Parameter | A number the model learns | 5000 and 2,00,000 |
| Hyperparameter | A setting you choose before training | The learning rate |
| Training | Adjusting parameters using data | Fitting on past sales |
| Inference | Using the trained model | Pricing a new house |
| Loss | How wrong the model is | Average squared error |
| Epoch | One full pass through the training data | All 10,000 sales seen once |
Common mistakes
- Mixing up parameters and hyperparameters. The model learns the first. You set the second.
- Mixing up the algorithm and the model. The algorithm makes the model, like a recipe makes a dish.
Check yourself
Summary
- Features go in, and labels are the right answers.
- An algorithm trains a model, and the model holds parameters.
- Training is learning, and inference is using.
Finished reading?