Module 08: CMA-ES Advanced & Natural Evolution Strategies - Restarts, Large-Scale, and Natural Gradients

Beyond vanilla CMA-ES: restart strategies for multimodal landscapes, separable and large-scale variants, and the elegant NES framework that connects ES to natural gradient descent and REINFORCE.

Learning Objectives

1. Implement IPOP-CMA-ES and BIPOP-CMA-ES restart strategies
2. Understand sep-CMA-ES and VD-CMA for high dimensions
3. Derive the NES gradient estimator (score function method)
4. Implement xNES and SNES from scratch
5. Understand fitness shaping and the connection to REINFORCE

Concept Explanation

Coming soon.

Code Examples

Coming soon.

Exercises

Coming soon.

Milestone Checklist

• Built IPOP-CMA-ES
• Can choose between full/separable/VD CMA-ES
• Implemented xNES from scratch
• Understand NES-REINFORCE connection