Viet-Anh on Software Logo

What is: Noisy Linear Layer?

SourceNoisy Networks for Exploration
Data SourceCC BY-SA -

A Noisy Linear Layer is a linear layer with parametric noise added to the weights. This induced stochasticity can be used in reinforcement learning networks for the agent's policy to aid efficient exploration. The parameters of the noise are learned with gradient descent along with any other remaining network weights. Factorized Gaussian noise is the type of noise usually employed.

The noisy linear layer takes the form:

y=(b+Wx)+(b_noisyϵb+(W_noisyϵw)x)y = \left(b + Wx\right) + \left(b\_{noisy}\odot\epsilon^{b}+\left(W\_{noisy}\odot\epsilon^{w}\right)x\right)

where ϵb\epsilon^{b} and ϵw\epsilon^{w} are random variables.