liveBook · Manning

Standard deviation

As explained in Applied Reinforcement Learning

Standard deviation is the square root of variance and measures the typical spread of a probability distribution.

In a Gaussian policy for continuous actions, the network outputs a mean μ and a standard deviation σ. The policy samples actions from Normal(μ, σ); σ controls the distribution’s spread and how strongly deviations from μ affect log-probability. section 9.3

Comparing sampled actions

With μ = 0.7 and σ = 0.2, an action of 0.82 is more probable than an action of 0.3. A larger σ makes the policy more tolerant of actions far from the mean; a smaller σ makes them more costly in log-probability. section 9.3

How Applied Reinforcement Learning covers it

How deep it goes
Explains how σ affects action likelihood and works through two sampled-action comparisons.
Examples
Compares actions 0.82 and 0.3 under a Gaussian policy with μ = 0.7 and σ = 0.2.
What you can do after
Understand how changing σ affects policy spread and the likelihood assigned to sampled actions.
How it connects
Connects standard deviation to Gaussian distributions, continuous action spaces, and policy gradients.

Also explained in

Compare how 18 books explain standard deviation →

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Applied Reinforcement Learning. The section links point to the parts of the book it draws on.