Establishing Quantified Uncertainty in Neural Networks, EQUi(Ne)2
EQUi(NE)2, or Establishing Quantified Uncertainty in NEural NEtworks, is an open-source python library and companion, typescript-based web application designed to enable the simple, straightforward addition of uncertainty quantification to neural networks with classification (label) outputs and concomitant visualization.
Deep neural networks (DNNs) for supervised labeling problems are known to produce accurate results on a wide variety of learning tasks. However, when accuracy is the only objective, DNNs frequently make overconfident predictions, and they also always make a label prediction regardless of whether or not the test data belongs to any known labels.
EQUINE was created to simplify two kinds of uncertainty quantification for supervised labeling problems:
- Calibrated probabilities for each predicted label
- An in-distribution score, indicating whether any of the model's known labels should be trusted.
Dive into our documentation examples to get started. Additionally, we provide a companion web application.