Thumbs up, Thumbs down

Year 8

Summary

Students train a model to recognize thumbs up or thumbs down (classification problem)

To perform this, you will need a GPU, the relevant code and a camera.

Suggested steps

1. Collect examples (>30 for each)
2. Train the model
3. Test

Discussion

• How many examples do you think we need?
• Notice that a probability is displayed when you are testing the model. How do you think the machine classifies it?
• What is an epoch? Train the model using 10 epochs.
• What is an adversarial example? What happens if we give the computer something completely different to what it’s expecting? Try it.

• In groups, get them to experiment to improve the performance of their model and document the changes they made.

Why is this relevant?

Students’ knowledge of Digital Technologies concepts will be developed further by understanding the importance of having plenty of data to work with. They will learn how data is represented as a probability and how this is used in a classification problem to organize and filter data, and model attributes of real world objects. We will contrast this directly with binary, and discuss how an activation function is used to get this effect. They will also go through the machine learning steps of collecting data, training a model and testing it.

Assessment

Check that they have > 30 examples for thumbs up AND thumbs down.

Check that they trained their model for 10 epochs.

The groups each demonstrate to the class, and explain how they tweaked their model to improve performance. Tally the number of correct responses by the computer. The group with the highest accuracy wins.

Curriculum links

Links with the Digital Technologies curriculum area

ADD Links with other curriculum areas

Eyes, nose, mouth

Year 8

Summary

Students train a machine to identify facial features screen  (classification problem) Similar to the other one but this time using regression rather than classification.

To perform this, you will need a GPU, the relevant code and a camera.

Suggested steps

1. Collect examples (>30 for each)
2. Train the model
3. Test

Discussion

• How many examples do you think we need?
• Collect data for left and right eye, nose, mouth separately, by clicking on the relevant facial feature. This stores it as a set of coordinates which it maps to the picture.
• What is an epoch? Train the model using 10 epochs.
• What is regression? How is this different to the previous classification problem?

• In groups, get them to experiment to improve the performance of their model and document the changes they made.

Why is this relevant?

Students’ knowledge of Digital Technologies concepts will be developed further by understanding the importance of having plenty of data to work with. They will learn how data such as a set of coordinates can be cleverly mapped to an image via a naming convention. They will understand that regression happens on a continuum. They will also go through the machine learning steps of collecting data, training a model and testing it.

Assessment

Check that they have > 30 examples for each facial feature.

Check that they trained their model for 10 epochs.

The groups each demonstrate to the class, and explain how they tweaked their model to improve performance. Tally the number of correct responses by the computer. The group with the highest accuracy wins.

Curriculum links

Links with the Digital Technologies curriculum area

ADD Links with other curriculum areas