Skip to content

A neural-network based image classifier that quantifies its uncertainty using Bayesian methods, as described in Kendall and Gal (2017)

License

Notifications You must be signed in to change notification settings

saurabhmathur96/uncertain-classifier

Repository files navigation

uncertain-classifier

An image classifier that quantifies its uncertainty using Bayesian methods as described in Kendall and Gal (2017)

Epistemic uncertainty

  • Model is confused between two or more classes, i.e. the image is a digit but it's not clear which one.
  • Can be fixed with more data.

Aleatoric uncertainty

  • Data is out-of-domain or the data is noisy.
  • Cannot be fixed with more data

Epistemic model

$ python3 train.py epistemic
$ python3 vis.py epistemic

Aleatoric model

$ python3 train.py aleatoric
$ python3 vis.py aleatoric

Combined model

$ python3 train.py combined
$ python3 vis.py combined

About

A neural-network based image classifier that quantifies its uncertainty using Bayesian methods, as described in Kendall and Gal (2017)

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages