Structures that react to sound.
The travelling salesman problem can be approximated with a self-organizing map by mapping 2D coordinates to a 1D modular arithmetic.
Generative adversarial networks (GANs) can be used to augment datasets for deep learning purposes: If there's not enough training data, just generate some more. The image below shows samples generated for training a ball/robot classifier.
View paper Google Colab Download dataset
A fresh take on Conway's Game of Life. But this time, it's in continuous space.