A Theory of Non-Acyclic Generative Flow Networks
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2023
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866913347550576640 |
|---|---|
| author | Brunswic, Leo Maxime Li, Yinchuan Xu, Yushun Jui, Shangling Ma, Lizhuang |
| author_facet | Brunswic, Leo Maxime Li, Yinchuan Xu, Yushun Jui, Shangling Ma, Lizhuang |
| contents | 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2312_15246 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | A Theory of Non-Acyclic Generative Flow Networks Brunswic, Leo Maxime Li, Yinchuan Xu, Yushun Jui, Shangling Ma, Lizhuang Machine Learning Artificial Intelligence Numerical Analysis Probability 68T07, 68T20, 60J05, 60J20, 60J22, 65F45, 65J20, 68T05, 68T20 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. |
| title | A Theory of Non-Acyclic Generative Flow Networks |
| topic | Machine Learning Artificial Intelligence Numerical Analysis Probability 68T07, 68T20, 60J05, 60J20, 60J22, 65F45, 65J20, 68T05, 68T20 |
| url | https://arxiv.org/abs/2312.15246 |