David Ha • 10/29/2017

A Visual Guide to Evolution Strategies

This article provides a visual introduction to Evolution Strategies (ES), a black-box optimization technique useful for problems where gradient calculation (like backpropagation) is difficult, such as in reinforcement learning with sparse rewards. It contrasts ES with traditional RL, highlights benefits like parallelization and policy diversity, and previews applications to tasks like MNIST and robotics environments.

0 comments

#optimization #Neural Networks #Reinforcement Learning