Deep Evidential Regression on Toy Dataset

[nilsleh]

Thank you for the interesting paper and publishing your code. In order to get a better understanding of your method, I have created a google colab (in pytorch) with a toy regression problem. I have implemented your methodology as well as the one from a follow up paper by Meinert et al. 2022. However, neither seem to perform very well, as especially the epistemic uncertainty is not behaving as expected (see below). Therefore, I was wondering if you could give me a hint about what I am missing/doing wrong or a possible explanation. Thanks in advance!