Creative Prediction with Neural Networks
A course in ML/AI for creative expression
Making Predictive NIMEs with Neural Networks
Charles Martin - The Australian National University
Outline
What kinds of neural networks are useful in NIMEs?
How do recurrent neural networks work?
How can we use them to generate creative data?
Is there a more convenient way to do this?
Outline
Focus on hands on experiments
Focus on small data
Use widely available tools
Everybody will train a neural network (?)
Rough Plan
- 9:00 Meeting, Intro, and Python Setup
- 9:15 Overview of Deep Learning and Creativity
- 9:45 Generating Text and Music with RNNs (practical)
- 10:30 (coffee break)
- 10:45 Hack: inventing Star Trek episode titles
- 11:30 Using Mixture Density Networks (MDNs)
- 13:00 Lunch (90 minutes)
- 14:30 Making Predictive Musical Interactions with the IMPS system
- 16:00 (coffee break)
- 16:15 Future directions for creative neural networks at NIME and beyond
Learning to Predict Sequences
Recurrent Neural Networks and LSTM Units
Mixture Density RNNs
Interacting with Musical Predictions
Starting a community
Predictions
Musical Predictions
What are predictions?
Where can they fit into a NIME?
Where can they fit into a performance?
Example: Wekinator-style NIMEs
Example: RoboJam
Example: Neural iPad Ensemble
talk
what kinds of predictions would you like a NIME to make?
Tools
Python is powerful.
We can all be Python superheroes!
Python setup can be hard.
(Sorry)
Colab to the Rescue?
extra tools
Or use whatever computer music environment you want (needs to send and receive OSC messages).
get your tools together
Now’s the time to look at the setup page on creativeprediction.xyz.
This course works best with Python, Keras, and Jupyter on your computer, but you can use (online) Google Colab Notebooks as well for most of the examples.