Skip to content
/ matNEAT Public

MATLAB implementation of Neuroevolution of Augmenting Topologies (NEAT)

License

Notifications You must be signed in to change notification settings

agaier/matNEAT

Repository files navigation

matNEAT

Cleaned up version of NEAT code used in the paper: "Data-Efficient Neuroevolution with Kernel-Based Surrogate Models" (https://arxiv.org/abs/1804.05364)

The surrogate-assistance/Gaussian process code has been stripped out for those interested only in a MATLAB version of NEAT. It includes experiments for XOR, double pole balancing, cart-pole swing up, and half-cheetah (with MuJoCo and OpenAI gym).

See usage examples in 'demo.m' for, um, examples of usage.

About

MATLAB implementation of Neuroevolution of Augmenting Topologies (NEAT)

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published