04_training

Training a Neural Network

Session A: Training a Neural Network

Objectives:

• Learn steps to construct a vanilla neural network and train a classification model with ml5.js.
• Understand Neural Network architecture
  • What is a Perceptron?
  • What is a multi-layered perceptron?
  • Activation Functions
• Understand the terminology of the training process
  • Training
  • Learning Rate
  • Epochs
  • Batch size
  • Loss
• Understand the difference between training and inference

Background: Perceptron and Artifical Neural Networks

• What is a Neural Network? by 3Blue1Brown
• Invented in 1957 by Frank Rosenblatt at the Cornell Aeronautical Laboratory (original paper), a perceptron is the simplest neural network possible: a computational model of a single neuron. A perceptron consists of one or more inputs, a processor, and a single output. (More to this history
• NOC Neural Network videos - 10.1 to 10.3 cover the "Perceptron", a model of a single neuron. The Perceptron forms the basis of modern multi-layer deep learning networks.
• NOC Neural Network chapter 10 - written explanation of Perceptron and accompanying code in 10.1 to 10.4.

p5.js Oscillators

• p5.js Oscillator example
• Sound Synthesis video tutorial

Related projects that map gesture to sound

• MARtLET by Michelle Nagai
• From the Waters by Anne Hege
• This Is Not A Theremin by Guillermo Montecinos and Sofía Suazo
• Eye Conductor by Andreas Refsgaard

Machine Learning for Human Creative Practice

• Machine Learning for Human Creative Practice, talk from Dr. Rebecca Fiebrink at Eyeo 2018.

ml5.js: Train Your Own Neural Network

• ml5.js: Train Your Own Neural Network (35 min)
• ml5.js: Save Neural Network Training Data (12 min)
• ml5.js: Save Neural Network Trained Model (11 min)
• ml5: Neural Network Regression (10 min)

Session B: Real-time Data

Objectives:

• Revisit and examine the concepts of classification and regression as applied to real-time interaction.

Pose Data as inputs to Neural Network

• 🎥 ml5.js: Pose Classification with PoseNet and ml5.neuralNetwork()
  • 💻 Collect Data
  • 💻 Load Data
  • 💻 Deploy Model
• 🎥 ml5.js: Pose Regression with PoseNet and ml5.neuralNetwork()
  • 💻 code all steps

Pixel Data as inputs to Neural Network

• 🎥 ml5.js: Train a Neural Network with Pixels as Input
  • 💻 simple image classifier code

Face Data

• 💻 Train regression model model with Face Keypoints

Project References

• ml5 Playful Examples by Andreas Refsgaard
• Creatability

Assignment 4 Due Sunday October 4

1. Watch Machine Learning for Human Creative Practice, Dr. Rebecca Fiebrink at Eyeo 2018. Write a response to the following question posed by Dr. Fiebrink:
   • How can machine learning support people's existing creative practices? Expand people's creative capabilities?
2. Dream up and design the inputs and outputs of a real-time machine learning system for interaction and audio/visual performance. This could be an idea well beyond the scope of what you can do in a weekly exercise.
3. Create your own p5+ml5 sketch that trains a model with real-time interactive data. This can be a prototype of the aforementioned idea or a simple exercise where you run this week's code examples with your own data. Here are some exercise suggestions:
   • Try to invent more elegant and intuitive interaction for collecting real-time data beyond clicking buttons?
   • What other real-time inputs might you consider beyond mouse position, image pixels, or face/pose tracking? Could you use real-time sensor data?
   • What other real-time outputs might you consider beyond color or sound modulation? Could the output be a physical computing device? Multiple outputs like R,G,B values?
4. Complete a blog post with your response, real-time ML system, and documentation of your code exercise and link from the homework wiki.