3 ms·
In what circumstances is it only 10x slower? Random search is totally useless when your environment is stochastic. These algorithms aren't learning sequences of
by cshenton 9y ago
In what circumstances is it only 10x slower? Random search is totally useless when your environment is stochastic. These algorithms aren't learning sequences of actions, in fact most use a 30 'no op' random start to avoid just that.
- backpropaganda 9y agoBy 'random', he means evolutionary search. It's not really random, and is just a slower method for policy gradient. Here's the OpenAI blog post: https://blog.openai.com/evolution-strategies/ https://blog.openai.com/evolution-strategies/