Skip to content

Homework 6

Jump to bottom Edit New page
lkkchung edited this page Apr 10, 2018 · 15 revisions

Neural Networks!

The assignment is to put your own creative "spin" on one of the examples (or create your own new example). Here are some suggestions but by no means do you need to stick to them.

  • Explore the ml5 library, can you the examples to run? Try incorporating image classification with a pre-trained model into a p5 sketch. If you are feeling ambitious, try building your own "teachable machine".
  • Follow the Doodle Classifier tutorials and consider adding one more more features:
    • Create a progress bar for training and testing using the draw() loop.
    • Add more training data (for additional categories)
    • Can you figure out a way to "save the model" as a JSON file for loading later?
    • What other image datasets could you use?
  • Continue the Color Predictor code from in class. Can you build out your own working, interactive version? For reference, here is Jabril's live stream explaining the color predictor as well as a a very silly video about it.
  • UPDATE I made a version of the Color Predictor: video tutorial, code, web editor code. Some things you might try:
    • Train the predictor to generate a "complementary" color.
    • Can you save the model for reloading later?
    • What other input / output color data might you use?
    • How can you make the interaction design more elegant?
  • Taking inspiration from the 3Blue1Brown series, can you visualize a neural network? (Even with just pretend data). This Network Animation example might help you get started.

Directions

  • Document your work on the web and link to it below.
  • Add any questions below:

Questions

  • How might I implement backpropagation? --Lucas
  • How do we know if this is working? --Lucas

Post your work

Add a custom footer
Add a custom sidebar
Clone this wiki locally