1 of 12

Engineering and Design Challenge – Teachable Machines

2 of 12

Engineering and Design Challenge

Lesson Target: I can design a machine to translate American Sign Language.

Success Criteria: Students will be able to teach their teachable machine sign language

Daily Tasks: Understand the engineering and design process and train our machines

ATL SKILL: Researchers

3 of 12

Engineering and Design Process

4 of 12

Step One: Ask

  • Define the Problem:

Can We build a teachable machine to recognize sign language?

  • Identify the need
  • Examine the evidence
  • Begin to form a claim

​

5 of 12

Step Two: Research

  • Look for background information
  • Understand the problem
  • Look for quantitative and qualitative information
  • Be informed as you formulate a solution

​

  • https://www.signasl.org/

6 of 12

Simple Sign Language Words

Will the computer recognize yes vs goodbye?

7 of 12

Step Three: Imagine

  • Brainstorm
  • Give and accept feedback of ideas while working positively with teammates
  • Begin to plan your solution – use your worksheet
  • Consider project criteria and constraints
  • How will you design your teachable machine to recognize ASL?

8 of 12

Step Four: Plan it!

  • Develop multiple ideas that will solve the problem
    • Criteria – Guidelines
      • Will the solution work?
      • What pictures do we need?
      • What will the result look like?
    • Constraints – Roadblocks
      • Deadlines
      • Sample size
      • Time to train the algorithm
  • Decide on an idea that best meets the criteria and fits within the constraints

9 of 12

Step Five: Create a Prototype

  • A prototype is a working model
  • looks and functions just like the finished product
  • Justify the chosen design based on the project criteria and identified constraints

​

10 of 12

Remember the Machine Learning Recipe��

  • Collect data [images of signs]
  • “Label” the data: mark what the “true” label of the data is which they want the machine to learn
  • The algorithm then trains a model to learn patterns and get better at predicting the label 
  • Give the algorithm a new data point without a label, and ask it to make a prediction of a label.

11 of 12

Step Six: Test and Evaluate

  • Test to find and discover problems
  • Answer these questions:
    • Is it safe for people and the environment?
    • Will it last as it needs to?
    • Does it meet the criteria and constraints?
    • Does it work?
  • Redesign – make changes as needed
  • Test new solutions

12 of 12

Step Seven: Improve Design � and Evaluate your Solution

  • After studying all test data and evaluating design solutions, you may need to make changes
  • This is the time to improve design before production starts
  • Document what improvements you are going to make on your worksheet