Normalizing Flows
Models
What are Normalizing Flows?
• Normalizing flows are a family of probabilistic models that can be used to
learn complex probability distributions.
• They work by transforming a simple, easy-to-sample distribution, such as
a Gaussian, into a more complex distribution by applying a series of
invertible transformations to the input distribution.
Deep Learning Part 6: Generative Modelling through Normalizing Flows written by Tejpal Kumwat ([Link]
through-normalizing-flows-c79fffc90091)
How Normalizing Flow work?
• Use a standard distribution, like a Gaussian ,which is easy to sample and
compute.
• Apply invertible Transformations by transforming the simple distribution into
the target complex distribution using a series of invertible.
Deep Learning Part 6: Generative Modelling through Normalizing Flows written by Tejpal Kumwat ([Link]
through-normalizing-flows-c79fffc90091)
Compute Data Density Using Change of Variables:
• The density of the data is computed as:
• The Jacobian determinant tracks how the transformation scales the probability density.
• Exact likelihood evaluation is used to evaluate flow based models.
• The path traversed by the random variables z is the flow and the full chain formed by the successive
transformations f is called a normalizing flow.
Deep Learning Part 6: Generative Modelling through Normalizing Flows written by Tejpal Kumwat ([Link]
modelling-through-normalizing-flows-c79fffc90091 )
Applications of Normalizing
Flows:
• Generative modeling: Normalizing flows can be used to generate new data that is similar to
the data that was used to train the model.
• Density estimation: Normalizing flows can be used to estimate the probability density of a
given data set.
• Inference: Normalizing flows can be used to perform inference on a given data set, such as
finding the most likely latent variables or the most likely parameters of the model.
Deep Learning Part 6: Generative Modelling through Normalizing Flows written by Tejpal Kumwat ([Link]
normalizing-flows-c79fffc90091)