Evolution Strategies
An evolutionary optimization method independently developed by Rechenberg and Schwefel. It requires no gradient information, naturally supports large-scale parallelism, and can serve as a scalable alternative to reinforcement learning.
Timeline (last 90 days)
Evolutionary search is a class of black-box optimization algorithms inspired by biological evolution, with typical representatives including genetic algorithms (GA), evolution strategies (ES), and CMA-ES.
A paper published by OpenAI in 2017 demonstrated that simple evolutionary strategy methods can match deep reinforcement learning on Atari games and MuJoCo control tasks, and inherently support large-scale parallelism.
All Facts (2)
A paper published by OpenAI in 2017 demonstrated that simple evolutionary strategy methods can match deep reinforcement learning on Atari games and MuJoCo control tasks, and inherently support large-scale parallelism.
60%UnverifiedEvolutionary search is a class of black-box optimization algorithms inspired by biological evolution, with typical representatives including genetic algorithms (GA), evolution strategies (ES), and CMA-ES.
50%