Teachable Machine
Overview: Working with simple hand-drawn shapes on sticky notes, participants will teach a computer to tell two shapes apart by capturing training photos, training a model, and testing it live with their camera. Along the way, they'll discover how the quantity and diversity of training data directly affect a model's accuracy, gaining an intuitive, firsthand understanding of how machine learning actually works and what it takes to make AI smarter.
Instructions:
1. Hand out 2 sticky notes and a dark washable marker.
2. Students decide on 2 distrinct shapes to draw. One shape per sticky note.
3. Tip: Simple shapes are better than detailed drawings (i.e. circles, star, square, triangle etc)
4. Go to teachablemachine.withgoogle.com/train
5. Select ‘Image Project’
6. Select ‘Standard Model’
7. Label Class 1 to <SHAPE 1> (this should reflect the name of the actual shape)
8. Take 20 photos of <SHAPE 1>
9. Label Class 2 to <SHAPE 2> (this should reflect the name of the actual shape)
10. Take 20 photos of <SHAPE 2>
11. Select ‘Train the model’
12. Test the model
- Present your sticky note to the camera with either Shape 1 or Shape 2 and see how confident it categorizes your shape!
Follow up questions for students:
How accurate is it? What can improve the model?
Follow Up Exercise (Optimizing the Model):
1. Take a total of 60+ photos for <SHAPE 1>
- Ensure photos are diverse and have different depths and angles of the shape.
2. Repeat for <SHAPE 2>
3. Select ‘Train the Model’
4. Test the model
Follow up questions for students:
Did the model performance improve? How does diversity of data change the outcome? How does the amount of data change the outcome? What happens when both stickies are off the screen? Does it improperly categorize this?
Follow Up Exercise (Optimizing the Model 2):
1. Take 20+ photos of a default (blank) background.
2. Retrain the model
3. Test the model
The model should have a 90%+ accuracy of categorizing the shapes and when there is no sticky note visible (default).
Stretch Assignment(9+ grade): Embedded Model for Arduino
This requires an Arduino and Ov7670 Camera
Sample Slide Deck: https://drive.google.com/file/d/1iVgttpwiMdcISPjarWipA-XeO-e8tyg3/view?usp=sharing