Biased Dreams: Limitations to Epistemic Uncertainty Quantification in Latent Space Models

This repository contains the training and evaluation code for the paper Biased Dreams: Limitations to Epistemic Uncertainty Quantification in Latent Space Models.

Structure

This repository is organized into two codebases: infoprop and uncertainty-aware-dreamer. We recommend setting up separate environments for each, as they may require conflicting package versions. Refer to infoprop/README.md and uncertainty-aware-dreamer/README.md for corresponding details on setup, experiment runs, and evaluation.

Citation

To cite our work, please use the following BibTeX entry:

@article{berger2026biaseddreams,
    title={{Biased Dreams: Limitations to Epistemic Uncertainty Quantification in Latent Space Models}},
    author={Julia Berger and Bernd Frauenknecht and Sebastian Trimpe and Bastian Leibe},
    journal={Reinforcement Learning Journal},
    year={2026},
    note={Presented at the Reinforcement Learning Conference (RLC 2026)}
}