MathieuDonofrio / enet

Fast lightweight neuro-evolution of augmenting topologies (neat) implementation

ENET

ENET is a small and simple header-only neuroevolution library based off Kenneth O. Stanley and Risto Miikkulainen's NEAT Paper.

Table of Contents

• Introduction

Introduction

Neuro Evolution of Augmenting Topogies (NEAT) is a fun way to evolve complexifying artifical neural networks. NEAT is often used to evolve small neural networks and is good at making interesting artifical life simulations. What NEAT teaches us:

• Innovation can be protected with speciation by using historical markings to track gene history.
• Weights and topology can be simultaneously optimized producing complexification over time that can be more optimal than gradient descent in certain cases.

This particular library simplifies certain operations to trade off efficiency for more compute allowing larger simulations to take place.

For further details:

• Paper: NEAT Paper