Master Assignment
[B] Medical OOD detection with an Exponentially Tilted Gaussian Prior for beta-Variational Autoencoders
Type: Master EE/CS/ITC
Period: TBD
Student: (Unassigned)
If you are interested please contact :
Background:
Exponentially Tilted Gaussian Priors [1] are shown to better model latent spaces in VAEs, however they have not been tried out on beta-VAEs. Beta-VAEs can provide disentangled latent spaces, which in turn can be used for OOD detection using likelihoods [2]. But likelihood ratio tests are shown to be better than likelihoods [3]. Your task is to test out LRTs on beta-VAEs modeled by Exponentially Tilted Gaussian Priors on benchmark medical image datasets and compare their performance with standard OOD detection methods.
Resources:
- The Exponentially Tilted Gaussian Prior for Variational Autoencoders https://arxiv.org/abs/2111.15646
- Unsupervised anomaly localization using VAE and beta-VAE https://arxiv.org/abs/2005.10686
- Likelihood Ratios for Out-of-Distribution Detection https://proceedings.neurips.cc/paper_files/paper/2019/file/1e79596878b2320cac26dd792a6c51c9-Paper.pdf
- MOOD 2020: A Public Benchmark for Out-of-Distribution Detection and Localization on Medical Images https://pubmed.ncbi.nlm.nih.gov/35468060/
- A Benchmark of Medical Out of Distribution Detection https://openreview.net/forum?id=oUg5rC_95OM
- MediCLIP: Adapting CLIP for Few-shot Medical Image Anomaly Detection https://papers.miccai.org/miccai-2024/504-Paper0333.html
- PyTorch-OOD: A Library for Out-of-Distribution Detection Based on PyTorch https://openaccess.thecvf.com/content/CVPR2022W/HCIS/html/Kirchheim_PyTorch-OOD_A_Library_for_Out-of-Distribution_Detection_Based_on_PyTorch_CVPRW_2022_paper.html