Neural Network Simulation in 3D Environments
Agent-based models (ABM) of evolutive artificial neural networks
This is a part of my project to simulate neural networks interactions in a 3D environments. The first layer of the network (yellow in the animation below) is the sensorial layer, where the inputs are identified to be processed by deeper layers.
The network starts with a symmetric topology and random connections. Using genetic algorithms, not only the weights are modified but also the structure of the network and the way that each neuron communicates with their neighbours. Other rules (e.g. activation functions and some stochastic processes) are also changed as the network evolves.
The goal is to use inspiration from theoretical neuroscience to implement new architectures and algorithms that can be used in next generation of machine intelligence.
“Better understanding biological brains could play a vital role in building intelligent machines.”
- Demis Hassabis