This repository is still under construction ... you're more than invited to try stuff, but strange things may happen ;)

chofer_torchex

In this repository I gather my extensions to PyTorch. The packaging structure tries to reproduce PyTorch's structure in order to facilitate usage for people familiar with PyTorch. In the following, you can find list of its main features (admittedly short ... yet ;) ).

1. nn.SLayer: This is an input layer which can operate on multisets of points in some cartesian product of the real numbers. Its primary intention is to train networks on the output of a topological data analysis pipeline, but can be used on arbitrary (real vector) multiset input. Tutorial

See Deep Learning with Topological Signatures for further reading.

@inproceedings{Hofer17c,
  author    = {C.~Hofer and R.~Kwitt and M.~Niethammer and A.~Uhl},
  title     = {Deep Learning with Topological Signatures},
  booktitle = {NIPS},
  year      = 2017,
  note      = {accepted}
}

The folder tutorials contains minimalistic examples in form of Jupyter notebooks to demonstrate how to use the PyTorch extensions.