About
an online ball game in which the trajectories of the moving balls are modulated by a real-time oscillating neural network. The neurons in the nerveBall network have continuous activation values that fluctuate between -inf and inf. The small scale (one neuron for each ball) network is single-layered and fully connected. The network is tuned by manipulating the activation and backpropagation algorithms. The network is an oscillating system, so it behaves like a musical instrument, and has to be tuned. This kind of network could be also used to generate musical data (like MIDI note events and parameter changes). Originally, the idea for the nerveBall network came from the research of Pauli Laine, an expert in composition algorithms, who called his model MINN in his doctoral thesis. Developing real-time oscillating networks is the dream which is why I got into neural network research in 2000. At the moment, in nerveBall, the network fluctuations are restricted and used to modulate the trajectories of the balls. nerveBall uses p5.js for video and audio output, user interaction and multitasking. nerveBall is released under GPL v3.
