Skip to yearly menu bar Skip to main content

In-Person Poster presentation / poster accept

GFlowNets and variational inference

Nikolay Malkin · Salem Lahlou · Tristan Deleu · Xu Ji · Edward Hu · Katie Everett · Dinghuai Zhang · Yoshua Bengio

MH1-2-3-4 #112

Keywords: [ Probabilistic Methods ] [ GFlowNets ] [ variational inference ] [ probabilistic modeling ] [ weighted importance sampling ]


This paper builds bridges between two families of probabilistic algorithms: (hierarchical) variational inference (VI), which is typically used to model distributions over continuous spaces, and generative flow networks (GFlowNets), which have been used for distributions over discrete structures such as graphs. We demonstrate that, in certain cases, VI algorithms are equivalent to special cases of GFlowNets in the sense of equality of expected gradients of their learning objectives. We then point out the differences between the two families and show how these differences emerge experimentally. Notably, GFlowNets, which borrow ideas from reinforcement learning, are more amenable than VI to off-policy training without the cost of high gradient variance induced by importance sampling. We argue that this property of GFlowNets can provide advantages for capturing diversity in multimodal target distributions. Code:

Chat is not available.