AI for absolute beginners
Learn what a model does before writing any code. By the end, you should be able to explain training, prediction and testing in your own words.
1. The problem comes first
Suppose a teacher wants to identify students who may need extra practice. The question is not “Which algorithm should we use?” but “What do we want to predict, and will it help students fairly?”
Inputs (features): practice exercises completed and quiz results. Output (target): whether extra support may be helpful. These are examples, not a recommendation to automatically judge students.
2. Examples help a model learn
A supervised learning model is trained using examples that contain both the inputs and the correct target. It finds a mathematical relationship between them. It may generalize imperfectly to new students.
3. Training is not testing
Set some examples aside before training. Test the finished model on this unseen set. A model that remembers training examples may still fail on new data. For classification, study precision and recall as well as overall accuracy.
4. Three main learning styles
Supervised
Learn from labelled examples: spam/not spam or estimated house price.
Unsupervised
Find structure without supplied target labels: grouping similar items.
Reinforcement
Learn actions through feedback and rewards in an environment.
Quick self-check
Try it yourself
- Think of a prediction you would like a computer to make.
- Write down the inputs you might collect. Ask whether you have permission to use them.
- Identify the target answer and how you would check whether predictions are correct.
- Consider who might be harmed if the model is wrong.