Machine Learning For Kids :: Teachers’ notes
Worksheet
Car or Cup
Activity Train the computer to be able to sort photos into groups.
Objective Teach a computer to recognise pictures of objects
• How computers can be trained to recognise pictures.
• The important of variety in training machine learning systems.
Difficulty level Beginner
Time estimate 45 minutes
Summary Students will train a machine learning model to recognise pictures of cars or cups. They
will use this to make a project in Scratch that sorts a pile of photos into groups.
Topics image classification, supervised learning
Setup
Each student will need:
Print-outs Project worksheet (download from https://machinelearningforkids.co.uk/worksheets )
Blocks in Scratch scripts are colour-coded, so printing in colour will make it easier for students.
Access Access to an image search site (e.g. Google Images, Bing Images, etc.)
Access Username and password for machinelearningforkids.co.uk
Class account will need:
API keys None
Customizing
If you use PRIMM approaches with your class, add a step where students predict how the project template works.
If you want to increase the amount of coding involved, delete some of the code from the project template and
add steps to the worksheet so students code it themselves.
If you want to encourage problem solving, delete some of the detail in the worksheets and provide more general
instructions instead.
Project template files & worksheets in MS Word format are available so you can modify them to suit your class.
Project https://github.com/IBM/taxinomitis-docs/tree/master/scratch-templates
templates Scratch 3 templates end .sb3 Scratch 2 templates end .sb2
Worksheets https://github.com/IBM/taxinomitis-docs/tree/master/project-worksheets/msword
Help
Potential • Students need Internet access to search for pictures of cars and cups to train the
issues computer with. Close supervision may be appropriate to ensure safe searching.
• The starter Scratch project includes a test set of images. Accuracy will be affected
by how similar these are to the students’ training images. For example, if students
collect examples of sports cars to train the computer to recognise cars, this may
struggle to recognise non-sports cars. If this happens, encourage them to think
about why it’s getting things wrong, and how they could improve this by collecting
a more varied set of photos to train the computer with.
• Dragging and dropping doesn’t work in Internet Explorer. You can provide your
students with a different web browser (Firefox or Chrome work well) or explain to
them how to copy/paste image URLs from a page.
• You cannot drag and drop pictures between different types of browser. In other
words, you can’t drag a picture from a Firefox window to Machine Learning for
Kids running in Chrome. Or from a Chrome window to Machine Learning for Kids
running in Firefox. You need to use the same type of web browser for both.
• Only jpg and png images can be used. If students find other image types (e.g. gif
files) and try to drag them into their training buckets, they’ll get an error. Explain
to them that this is okay, and they should just choose a different picture.
• If there is a problem with any of the pictures they find, they’ll be shown as a red
box. They can click on the cross for these broken images to remove them.
General troubleshooting and help at https://machinelearningforkids.co.uk/help
