Learning Materials for Evolution Strategies

私達の研究トピックの一つに，進化戦略（Evolution Strategy, ES）があります．進化戦略は目的関数の勾配を用いない連続パラメータの最適化法の一つです．とりわけ，CMA-ES（Covariance Matrix Adaptation Evolution Strategy）は，汎用的で効率的な探索法として，様々な場面で利用されています．ここでは，CMA-ESを学ぶためのコンテンツを公開します．

One of our research topics is Evolution Strategy (ES). It is an optimization method for continuous parameters that does not rely on the gradient of the objective function. In particular, the Covariance Matrix Adaptation Evolution Strategy (CMA-ES) is used in a wide range of applications as a versatile and efficient search method. Here, we are releasing content to help you learn about CMA-ES.

What is Black Box Optimization?

ここでは，Black-Box 最適化を実際に自分の手で行うことで，ブラックボックス最適化とはどのようなものか，理解します．Black-Box最適化というものを知らない方が対象です．

ブラックボックス最適化とは[Open in Colab] [Open in GitHub]

This guide is for those who are new to black-box optimization. You will gain a hands-on understanding of what black-box optimization is by performing an example yourself.

What is Black Box Optimization? [Open in Colab] [Open in GitHub]

Understanding CMA-ES

ここでは，CMA-ESの各コンポーネントがなぜ必要なのか，いつ必要なのか，を理解するために，単純なランダムサーチからスタートし，ステップサイズ適応，分散適応，共分散行列適応，などの構成要素を一つずつ追加していった場合の効果を，実習を通して学びます．

Evolution Strategy [Open in Colab] [Open in GitHub]
ステップサイズ適応 [Open in Colab] [Open in GitHub]
分散適応（Separable CMA-ES） [Open in Colab] [Open in GitHub]
変数非分離性と共分散適応 [Open in Colab] [Open in GitHub]
多峰性関数の最適化 [Open in Colab] [Open in GitHub]
アドバンストな共分散適応メカニズム[Open in Colab] [Open in GitHub]
CSAの課題と別のステップサイズ適応方法（TPA）（更新予定）
高次元最適化のためのCMA-ES（更新予定）

（コンテンツは修正中．随時更新していきます．）

In this section, we'll start with a simple random search and, through hands-on exercises, progressively add components like step-size adaptation, variance adaptation, and covariance matrix adaptation. This will help us understand why and when each component of the CMA-ES is needed.

Evolution Strategy [Open in Colab] [Open in GitHub]
Step-size Adaptation [Open in Colab] [Open in GitHub]
Variance Adaptation (Separable CMA-ES) [Open in Colab] [Open in GitHub]
Variable Non-Separability and Covariance Matrix Adaptation [Open in Colab] [Open in GitHub]
Optimization of Multimodal Functions [Open in Colab] [Open in GitHub]
Advanced Covariance Matrix Adaptation Mechanisms [Open in Colab] [Open in GitHub]
Challenges of CSA and Another Step-size Adaptation Mechanism (TPA）（TBA）
CMA-ES for High-Dimensional Optimization (TBA)

Using CMA-ES

ここでは，CMA-ESをうまく利用するための実践ガイドを提供します．これからCMA-ESを自分の問題に対して利用したい方，CMA-ESを利用して得られた結果の解釈に困っている方，すでにCMA-ESを利用しているが望ましい結果が得られていない方，などが主な対象です．

CMA-ES実践ガイド [Open in Colab] [Open in GitHub]
CMA-ESによる最適化の高速化（更新予定）

This section provides a practical guide for effectively using CMA-ES. It's primarily intended for those who want to apply CMA-ES to their own problems, those who have difficulty interpreting results from its use, or those who are already using it but are not achieving the desired outcomes.

Practical Guide to CMA-ES [Open in Colab] [Open in GitHub]
Accelerating Optimization with CMA-ES (TBA)

Advanced Use of CMA-ES

ここでは，CMA-ESを発展的な最適化問題へと応用する方法を紹介します．

CMA-ESを用いたミニマックス最適化（最悪性能の最適化） [Open in Colab] [Open in GitHub]

This section introduces how to apply CMA-ES to advanced optimization problems.

Minimax Optimization with CMA-ES (Worst-Case Performance Optimization) [Open in Colab] [Open in GitHub]

Report abuse