Trajectory balance: Improved credit assignment in GFlowNets
This addresses a bottleneck in GFlowNets for researchers in generative modeling, though it is an incremental improvement over existing methods.
The paper tackles inefficient credit propagation in GFlowNets for generating compositional objects by proposing a new trajectory balance objective, which experimentally improves convergence, sample diversity, and robustness across four domains.
Generative flow networks (GFlowNets) are a method for learning a stochastic policy for generating compositional objects, such as graphs or strings, from a given unnormalized density by sequences of actions, where many possible action sequences may lead to the same object. We find previously proposed learning objectives for GFlowNets, flow matching and detailed balance, which are analogous to temporal difference learning, to be prone to inefficient credit propagation across long action sequences. We thus propose a new learning objective for GFlowNets, trajectory balance, as a more efficient alternative to previously used objectives. We prove that any global minimizer of the trajectory balance objective can define a policy that samples exactly from the target distribution. In experiments on four distinct domains, we empirically demonstrate the benefits of the trajectory balance objective for GFlowNet convergence, diversity of generated samples, and robustness to long action sequences and large action spaces.