Types of machine learning and their applications in the real world
Describe the types of machine learning and their applications in the real world
In this lesson
- Describing how machine learning differs from writing rules by hand
- Describing supervised, unsupervised, reinforcement, deep and transfer learning and what makes each one distinctive
- Choosing a suitable type of machine learning for a real-world application and justifying the choice
LearnLearning from data
LearnSupervised and unsupervised learning
LabSame data, two kinds of learning
Customers of a fitness app, by hours in the app and monthly spend. Switch between the labelled and unlabelled views.
Each training example is labelled: ● renewed, ✕ did not renew. Add a new customer to see the model predict a label for them.
Classify a new customer in the supervised view and run the clustering to the end in the unsupervised view to complete this activity.
LearnReinforcement learning
LabTrain an agent
The agent starts in the bottom-left corner. Moves cost 1 point; pits end the episode with a penalty; the two rewards end it too.
Episodes so far: 0. Each move costs -1; an episode ends at a reward or a pit.
The agent starts knowing nothing. Train it, then show the route it has learnt (only the best known moves, no exploring).
Simplified Q-learning. To complete this activity: train at least 200 episodes with no exploration and show the learnt route; then start again, train with some exploration, and show a route that reaches the +20 goal.
LearnDeep learning and transfer learning
SummaryReal-world applications
| Application | Typical approach | Why |
|---|---|---|
| Market basket analysis | Unsupervised (association) | Finds items bought together; there are no labels to predict |
| Medical imaging diagnostics | Supervised deep learning (CNNs) | Scans labelled with a diagnosis by doctors train the model to recognise signs of disease |
| Natural language processing | Deep learning (transformers), with transfer learning | Language is complex and sequential; large pre-trained language models are fine-tuned for each task |
| Object detection and classification | Supervised deep learning (CNNs) | Learns from images in which objects have been labelled |
| Robotics navigation | Reinforcement learning | The robot learns from rewards for reaching targets and penalties for collisions |
| Sentiment analysis | Supervised (deep) learning | Texts labelled positive, negative or neutral train the classifier |
CheckWhich type of learning?
Choose the type of machine learning that best describes each system.
- Predicting a used car's price from thousands of past sales whose prices are known
- Grouping news articles into topics that nobody has defined in advance
- A program that learns to play a video game by scoring points and losing lives
- Adapting a language model trained on billions of web pages to answer questions about one company's product manuals
CheckQuick check
Which statement about deep learning is correct?
ExamWorked exam answer
A supermarket analyses millions of receipts to find products that are often bought together. Describe why this is an example of unsupervised learning.
[2]ReviewMistakes examiners see
Saying reinforcement learning uses labelled data. It learns from rewards and penalties received after its own actions.
Describing transfer learning as copying a model unchanged to a new task. The pre-trained model is adapted to the new task (fine-tuned, or given a new output layer) using data for that task.
For “Describe” questions on types of ML, give the defining feature (labels, rewards, layers, reuse) and connect it to the application in the question.
Exam practiceExam-style questions
4 questions · 10 marks. Answer each one first, then mark it yourself against the markscheme. Your answers are saved in this browser only.
An agricultural technology company uses machine learning in three ways: drones photograph crops so that diseased plants can be detected; soil sensors in hundreds of fields send readings that are analysed to find fields with similar conditions; and a self-driving tractor learns to drive along crop rows without damaging plants.
Describe how supervised learning could be used to detect diseased plants in the drone photographs.
Describe why unsupervised learning is suitable for analysing the soil sensor readings.
Describe how reinforcement learning could train the tractor to drive along the crop rows.
The company has only 2000 photographs of plants with a rare crop disease. Describe how transfer learning could help it build an accurate detector.
0 of 4 questions marked. Mark every question to see your total.
Exam-style questions written for Praxis IB in the format of the 2027 specimen papers. They are not IB past-paper questions.