Challenging Euclidean Topological Autoencoders

This is a follow-up project of the ICML 2020 paper "Topological Autoencoders" (reference below). Here, we investigate whether domain-specific distance functions in the input space (here image datasets) are necessary for TopoAE, or whether a generic euclidean distance is sufficient. This work has been accepted for presentation at the Neurips 2020 TDA and Beyond workshop.

References

Please use the following BibTex code to cite our Neurips 2020 workshop paper:

@InProceedings{moor2020challenging,
    title       = {Challenging Euclidean Topological Autoencoders},
    author      = {Moor, Michael and Horn, Max and Borgwardt, Karsten and Rieck, Bastian},
    booktitle   = {NeurIPS 2020 Workshop on Topological Data Analysis and Beyond},
    year        = {2020},
    url         = {https://openreview.net/forum?id=P3dZuOUnyEY},
}

Furthermore, the original ICML 2020 paper proposing Topological Autoencoders in the first place, can be cited as follows:

@InProceedings{Moor19Topological,
  author        = {Moor, Michael and Horn, Max and Rieck, Bastian and Borgwardt, Karsten},
  title         = {Topological Autoencoders},
  year          = {2020},
  eprint        = {1906.00722},
  archiveprefix = {arXiv},
  primaryclass  = {cs.LG},
  booktitle     = {Proceedings of the 37th International Conference on Machine Learning~(ICML)},
  series        = {Proceedings of Machine Learning Research},
  publisher     = {PMLR},
  pubstate      = {forthcoming},
}

Setup

In order to reproduce the results indicated in the workshop paper simply setup an environment using poetry:

poetry install  

Running the methods:

Make sure you have internet access once to be able to download the datasets, and also the vgg model (via the lpips package)

In case a slurm cluster is available, simply run:

source scripts/run_slurm.sh  

Alternatively, all jobs can be sequentially/manually called using:

source scripts/run_manual.sh