Likelihood-free Out-of-Distribution Detection with Invertible Generative Models

Authors: Amirhossein Ahmadian, Fredrik Lindsten

IJCAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The performance results in terms of the Area Under ROC curve (AUROC) are given in Table 1 . We have experimented on some of the image dataset pairs which are common in the literature of OOD detection, such as MNIST vs. Fashion-MNIST, and CIFAR10 vs. SVHN.
Researcher Affiliation Academia Amirhossein Ahmadian , Fredrik Lindsten Division of Statistics and Machine Learning, Department of Computer and Information Science, Link oping University {amirhossein.ahmadian, fredrik.lindsten}@liu.se
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes The code and supplementary material are available at: https://github.com/aahmadian-liu/ood-likefree-invertible
Open Datasets Yes We have experimented on some of the image dataset pairs which are common in the literature of OOD detection, such as MNIST vs. Fashion-MNIST, and CIFAR10 vs. SVHN.
Dataset Splits Yes More specifically, the generative model is (or has been) trained on the entire standard training partition of the in-distribution data; the OSVM model is trained on 3000 random samples from the standard test partition of the in-distribution data, and is tested on 3000 different random samples from the test partition of this dataset in addition to 3000 random samples from the OOD dataset.
Hardware Specification Yes on a Titan X Pascal GPU with batch size=10.
Software Dependencies No The paper mentions that the models are 'implemented in Py Torch' but does not specify version numbers for PyTorch or any other software dependencies.
Experiment Setup Yes The settings for this approximation in our experiments can be found in appendix B. [...] on a Titan X Pascal GPU with batch size=10.