Remember:
Healthy predictions need healthy data!
Healthy
dataset
Finds
correct
patterns
Healthy
prediction!
Correct actions or decisions!
Do you remember what makes a healthy dataset?
- Lots of data
- Accurate
- Matches your problem and solution
- Different examples of data
- The right kind of data
- You have permission to use it
IT'S TIME FOR YOU TO START TRAINING YOUR DATA!
You should have gathered your data by now… or at least started!
WHETHER IT’S…
Your own training data from your community
and/or
Data gathered from sensors or user input
and/or
Data collected from public datasets
Click on a platform name below to learn more and try out some tutorials.
- Train images, sounds or poses
Attach devices to capture sensor data
Here are three tutorials to try out Teachable Machine using different data types.
- Image Classification with bananas
- Sound Classification with snaps, claps, and whistles
- Pose classification to detect head tilt
- Train images, sounds, text, or numbers
- Make a Scratch or Python project
- Or a mobile app with App Inventor
This video is an example of a Technovation team using Machine Learning for Kids to make a mobile app to sort biomedical waste.
TRY IT YOURSELF!
In this video, see how to take the Iris public dataset from Unit 6 and train it using Machine Learning for Kids.
Train images, sounds, or poses- With your trained dataset you can make a mobile app that uses AI.
WANT TO TRY IT?
In this video, see how you can use the App Inventor image classifier to train an image dataset. This dataset classifies healthy fruit vs diseased fruit.
ACTIVITY: TRAIN YOUR MODEL
- Choose the AI tool you want to use for your Technovation Project.
- Add your examples.
- Train and test your AI model.
Mentor Tip
Best practices: Training models is hard! Even Google gets it wrong. Their AI was trained but still started outputting wrong results! Don’t give up!
Guiding Questions to ask students: How accurate do you want your AI model to be? If it can not be 100% accurate, what is an acceptable answer? 80% of the time? Does that depend on the risk of what you are using the model for? For example self driving cars have to be pretty accurate otherwise they might hurt someone but google search results apparently have a much lower bar.
Mentor tips are provided by support from AmeriCorps.
