Frequentist bounds for multi-class classification

This repository contains an algorithm for multi-class classification that includes frequentist uncertainty intervals. As an example, we here apply the algorithm to the MNIST dataset. A description of the method can be found in:

• Dominik Baumann and Thomas B. Schön, "Safe reinforcement learning in uncertain contexts," IEEE Transactions on Robotics, 2024, arXiv.

Requirements

Code was developed using Python 3.8.10. The following libraries are required:

• numpy (developed with version 1.21.0)
• scikit-learn (developed with version 1.0.2)
• torchvision (developed with version 0.7.0)
• matplotlib (developed with version 3.3.4)

Execution

To execute the code, run

python main.py