Gang Fang's Blog

One more question on this statement:
“Importantly, when the reward function R represents the product of a prior (over some random variable) times a likelihood (measuring how well that choice of value of the random variable fits some data), the GFlowNet will learn to sample from the corresponding Bayesian posterior.”

What does the random variable refer to in the statement?

My answer: I don’t think there is a definition made in the article nor it can be easily specified. But generally speaking, in the context of Bayesian inference, the model parameters are taken as random variables. So it could be it.

But I think the critical insight provided by this statement is that GFlowNet learns the parameters of the model and will sample from the Bayesian posterior, implying it uses a Bayesian framework. And there will be more explanations in the later part of the article which might clarify this statement.