1 of 19

Dancing with AI: Day 2

2 of 19

Overview

​

  1. Teachable Machine Image vs. Pose Models Activity
  2. Discussion: Data Representation
  3. Data Bias and Privacy
  4. Brainstorming with Scratch + Teachable Machine
  5. Wrap-up

​

3 of 19

Teachable Machine! �

https://teachablemachine.withgoogle.com/

​

​

via Google Teachable Machine

4 of 19

Training a Machine Learning Model

Training

Input Samples

Learning Algorithm

Output

5 of 19

Let’s train a model together!

We can use Teachable Machine to train a model with image data, audio data, or pose data. In our class, we’ll learn to use images and poses as the input data to our AI systems.

To capture this data, the computer perceives using a camera.

6 of 19

What do you notice about these two examples?

​

���Image Model

���Pose Model

What happens with a different background?

​

​

How would you break the model?

​

​

Other observations?

Write down your thoughts in your journal!

7 of 19

Your turn!

  • On your laptops, navigate to https://teachablemachine.withgoogle.com/train
  • Come up with 2-3 body positions and train an image model and a pose model that can differentiate between them. Train at least 30 samples for each class.

​

Test it out! �What works well, and what doesn’t? Evaluate your models in your journal.

8 of 19

Awesome Job! You guys not only created machine learning models but also played around with...

Data Representation

9 of 19

Let’s Reflect!

  1. What are the pros and cons of using images versus poses?

​

  1. What are the differences between images and poses?

Data Representation can directly affect what �features and patterns the model learns while training.

10 of 19

Let’s talk about ETHICS...

Artificial Intelligence gives us a world of possibilities -- we can train models to learn on numerous types of data and apply those models to help solve real human problems.

But what are some of the downsides of machine learning models? What might go wrong?

11 of 19

<a href="https://www.freepik.com/free-photos-vectors/school">School photo created by freepik - www.freepik.com</a>

Bias

12 of 19

Bias

<a href="https://www.freepik.com/free-photos-vectors/school">School photo created by freepik - www.freepik.com</a>

13 of 19

Bias

AI systems can only learn from the data that they’ve been trained on. If AI is trained on incomplete, inaccurate, or biased datasets, it can have serious consequences down the road.

14 of 19

Let’s Reflect!

How would you make image/pose recognition systems less prone to bias?��

​

​

15 of 19

Privacy

How many of you would be okay with a hacker having access to your...

  • Email address
  • TikTok username
  • Phone number
  • Home address
  • A picture of your face

16 of 19

Privacy

​

What do you think

is okay to share

on the Internet?

​

​

17 of 19

Brainstorming a Project!

18 of 19

Think of a project that you could make!

In your journal, identify the three parts of your AI system:

  • What data does your system use?
  • What is the algorithm learning?
  • What is the output of your system?

Example Projects

Coding Cards

Be creative! Use the available resources if you need some inspiration.

19 of 19

Great Job!

That’s it for Day 2!

​

See you tomorrow, when we will learn more about block-based programming with poses.

CREDITS: This presentation template was created by Slidesgo, including icons by Flaticon, and infographics & images by Freepik.

​