Functional Flow Matching

This repo contains the code for our paper Functional Flow Matching (Oral, AISTATS 2024).

Our code is roughly structured as follows:

• /configs contains example configuration files that can be used to configure parameters for experiments
• /data contains the data used in our experiments
• /models contains the various neural architectures used in this work
• /scripts has files that can be used to launch training jobs to reproduce our experiments
• /util contains various utilities, e.g. for reading config files and performing evaluation
• The files functional_fm.py and conditional_ffm.py implement our FFM model.
• The file diffusion.py and losses.py implement the baseline DDPM and DDO models. Similarly, gano.py and gano1d.py implement the GANO baseline.

Citation

If you found our code useful or build upon our work, we ask that you cite our AISTATS 2024 paper as follows:

@inproceedings{kerrigan2024functional,
  title = {Functional FLow Matching,
  author = {Gavin Kerrigan and Giosue Migliorini and Padhraic Smyth},
  booktitle = {The 27th International Conference on AI and Statistics (AISTATS)},
  year = {2024}
}