The technique is inspired by biological evolution and speeds up the selection of machine learning algorithms.
Engineers at Waymo, owned by Alphabet, have teamed up with researchers at DeepMind, to build a technique that involves StarCraft II strategy game to help train neural networks for self-driving cars.
The researchers used the population-based training (PBT) technique which was previously used to train DeepMind to play StarCraft II. The technique is inspired by biological evolution and speeds up the selection of machine learning algorithms, the MIT Technology Review notes.
By using the technique, the AI algorithms guiding Waymo's autonomous cars can be retrained and recalibrated, accelerating the development of the AI algorithms in the hyper-competitive world of automated driving.