Class Notes

• nature of code chapter 10 (Neural Networks)
• OpenAI model spec
• perceptron: single neuron, made of 1) weighted sum and 2) an activation function
  • always has a bias and a bias weight in every layer
  • perceptron on its own was limited; with multiple layers made a breakthrough
• outputs are # of answer sets (because it’s likelihood of each)

Learning Materials

• Machine Learning for Human Creative Practice

A Brief History of Neural Nets and Deep Learning

Assignment Log

<aside> 💭

How can machine learning support and expand creative practices?

</aside>

• think less of what is possible and think of what ‘can be’ → similar to Dr.Fiebrink’s point on prototyping

<aside> 💡

Idea for a creative application that requires training a custom model?

</aside>

1. Aww face predictor

• I think the ‘aww face’ has some universality.

  

  • your forehead bunches up and your mouth sort of pouts.

• after training on the facial muscle movements on a calm face → an aww face, I believe a machine can ‘predict’ your aww face.

  • it would be a regression task.
  • training data can be collected from streamer data/youtube.