Distributional-accuracy diagnostics for GFlowNets

Establish principled methods, potentially inspired by the Markov chain Monte Carlo literature, for properly assessing the distributional accuracy of GFlowNets.

Background

The paper notes that diagnosing GFlowNets is difficult, particularly when assessing whether the learned terminal-state distribution accurately matches the target distribution. It asks whether diagnostics developed for Markov chain Monte Carlo can provide a reliable assessment framework for GFlowNet distributional accuracy.

References

As noted in , diagnosing GFlowNets is % a strikingly difficult; % problem; can we draw inspirations from the MCMC literature to properly assess the distributional accuracy of GFlowNets?

— Particle GFlowNets: Rethinking Generative Marginalization Models  (2609.11538 - Silva et al., 10 Sep 2026) in Section 6, Discussion