Skip to content

Latest commit

 

History

History
16 lines (13 loc) · 987 Bytes

README.md

File metadata and controls

16 lines (13 loc) · 987 Bytes

#Unsupervised DeepView for Uncertainty Visualization

This is an implementation of a novel framework allowing for unsupervised uncertainty visualization depicting a global decision boundary. Inspired by Deepview, it no longer requires labels for reliable visualization of misclassifications or adversarials. It solves the lack of global explainability methodologies for unsupervised learning areas that depict uncertain areas in data sets, whereas Deepview explicitly required labeled data to spot any uncertainties in the visualization method.

Key Features

  • model-agnostic, only requires prediction uncertanties and the raw data
  • global visualization method

Requirements

All required python libraries are stored in the requirements.txt file. Tested on Python 3.6.

pip install -r requirements.txt

Output of our visualization method

See DeepView_unsup.png