Stable GFlowNets with Probabilistic Guarantees
For practitioners using GFlowNets, this work addresses training instability and provides theoretical guarantees, though the improvement is incremental over existing methods.
GFlowNets suffer from training instability, including loss spikes and mode collapse. The authors derive loss-to-TV bounds to certify global fidelity and propose Stable GFlowNets, which empirically improve training stability and distributional fidelity.
Generative Flow Networks (GFlowNets) learn to sample states proportional to an unnormalized reward. Despite their theoretical promise, practical training is often unstable, exhibiting severe loss spikes and mode collapse. To tackle this, we first assess the sensitivity of GFlowNet objectives, demonstrating that a small Total Variation (TV) distance between the learned and target distributions does not preclude unbounded training loss. Motivated by this mismatch, we establish converse guarantees by deriving loss-to-TV bounds that certify global fidelity from bounded trajectory balance losses. Lastly, we propose Stable GFlowNets, an algorithm that leverages our theoretical results to stabilize training, and empirically demonstrate improved training behavior and superior distributional fidelity.