liveBook · Manning

Normal distribution

As explained in Applied Reinforcement Learning

A Gaussian distribution is a continuous probability distribution characterized by a mean and standard deviation.

A normal or Gaussian distribution describes uncertainty in a continuous variable using a mean and a standard deviation. It can model aggregate variability arising from many small, independent sources, such as variation in production time. section 2.5

In production planning, probability distributions represent uncertainty that affects transitions between states. A normal distribution is useful for continuous quantities; a Poisson distribution instead models counts of discrete events, such as urgent orders. section 2.5

Figure 2.6 Two important probability distributions widely used for modeling business processes.

Compare the visual forms of the normal and Poisson distributions used to model business processes. section 2.5

A Gaussian distribution can also represent a policy over continuous actions. A neural network produces the mean and standard deviation for the current state, and the agent samples an action from that distribution. REINFORCE then adjusts the policy using the sampled action and its return-to-go. section 9.3

Sampling a fan speed

At a temperature of 25°C and rising, a policy outputs a mean fan speed of 0.7 and a standard deviation of 0.2, then samples 0.82. If the return-to-go is −50, the update pushes the policy to make 0.82 less likely in that state. A positive return would instead reinforce the sampled action. section 9.3

How Applied Reinforcement Learning covers it

How deep it goes
Defines its parameters and illustrates their role in a policy update without a full numerical Gaussian-gradient calculation.
Examples
Uses variable production time and a sampled fan speed with a negative return-to-go.
What you can do after
Evaluate/understand when a Gaussian distribution can model continuous uncertainty or support a sampled-action policy.
How it connects
Connects to Poisson distribution, Markov decision processes, policy gradients, and REINFORCE.

Also explained in

Compare how 19 books explain normal distribution →

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.