Neuroevolution

Published:

Neuroevolution is a way of training neural networks by improving them gradually, rather than adjusting weights step by step with backpropagation. The system creates many different networks, tests how well each one performs, keeps the better versions, and then produces new networks by slightly modifying them.

Some neuroevolution approaches only evolve the network’s weights, while others also evolve the architecture itself. This means the system can add or remove connections as it searches for a better design. A well-known method called NEAT evolves both structure and weights together. Neuroevolution is often used in games and simulations where feedback is sparse or where exploring diverse strategies is more important than fine-tuning one solution.

Follow us on Facebook and LinkedIn to keep abreast of our latest news and articles