A Theory of Non-Acyclic Generative Flow Networks
Abstract
GFlowNets is a novel flow-based method for learning a stochastic policy to generate objects via a sequence of actions and with probability proportional to a given positive reward. We contribute to relaxing hypotheses limiting the application range of GFlowNets, in particular: acyclicity (or lack thereof). To this end, we extend the theory of GFlowNets on measurable spaces which includes continuous state spaces without cycle restrictions, and provide a generalization of cycles in this generalized context. We show that losses used so far push flows to get stuck into cycles and we define a family of losses solving this issue. Experiments on graphs and continuous tasks validate those principles.
Cite
Text
Brunswic et al. "A Theory of Non-Acyclic Generative Flow Networks." AAAI Conference on Artificial Intelligence, 2024. doi:10.1609/AAAI.V38I10.28989Markdown
[Brunswic et al. "A Theory of Non-Acyclic Generative Flow Networks." AAAI Conference on Artificial Intelligence, 2024.](https://mlanthology.org/aaai/2024/brunswic2024aaai-theory/) doi:10.1609/AAAI.V38I10.28989BibTeX
@inproceedings{brunswic2024aaai-theory,
title = {{A Theory of Non-Acyclic Generative Flow Networks}},
author = {Brunswic, Leo Maxime and Li, Yinchuan and Xu, Yushun and Feng, Yijun and Jui, Shangling and Ma, Lizhuang},
booktitle = {AAAI Conference on Artificial Intelligence},
year = {2024},
pages = {11124-11131},
doi = {10.1609/AAAI.V38I10.28989},
url = {https://mlanthology.org/aaai/2024/brunswic2024aaai-theory/}
}