進化戦略（ES）は一種の 最適化アルゴリズム that mimic the process of natural evolution to solve complex optimization problems. They are particularly effective in scenarios where traditional gradient-based methods may struggle, such as in high-dimensional spaces or when dealing with noisy or discontinuous objective functions.
At the core of Evolution Strategies is the concept of a population of candidate solutions, which are iteratively improved through selection, mutation, and recombination. In each iteration, a set of candidate solutions is generated based on the current population. These candidates are evaluated using a predefined 適応度関数, which measures their performance relative to the 最適化目標.
最も性能の良い候補が新しい集団を形成するために選択され、成功しなかった候補は破棄されます。選択された候補は突然変異を受け、小さなランダムな変化が導入されて多様性が生まれます。このプロセスにより、進化戦略は局所最適に陥ることなく、広範な解の探索を可能にします。
進化戦略には、(1+1)-ES、(μ/ρ, λ)-ESなどいくつかのバリアントがあり、選択、交叉、突然変異の扱い方が異なります。特に、ESは候補の性能に基づいて突然変異率を適応させることもでき、最適解を見つける能力をさらに高めます。
In the context of artificial intelligence and machine learning, Evolution Strategies have been successfully applied to tasks such as ニューラルアーキテクチャ探索, reinforcement learning, and optimizing hyperparameters. Their ability to operate in challenging optimization landscapes makes them a valuable tool in the AI practitioner’s toolkit.