technique

Neuroevolution

Evolve a population of neural networks with a genetic algorithm instead of training one network by backpropagation: fitness accumulates from behavior, selection is fitness-weighted.

What it is

How it works

Parameters & tuning

Where it’s been used

Variants & neighbours

Go deeper

Related, in brief

Covered here rather than as pages of their own — each is described by a single source, and gets promoted the moment a second one corroborates it.

NEAT (NeuroEvolution of Augmenting Topologies)

A neuroevolution method that evolves a neural network's structure (topology) as well as its weights, starting from simple networks and progressively complexifying them.

Whisker (ray-based) proximity sensors

Give an agent several ray-like sensors radiating from its center that measure distance to nearby obstacles, feeding those distances into its neural network as inputs.

Connected to

Built from: Perceptron

Further reading: The Nature of Code

Kinds of this: NEAT (NeuroEvolution of Augmenting Topologies)

Contrast with: Roulette wheel selection

Used to build: Whisker (ray-based) proximity sensors

Revision 1 · 1 source · Something wrong? Tell us.