Learning to Scale Logits for Temperature-Conditional GFlowNets
This work addresses a specific bottleneck in training controllable generative models for researchers in computational biology and chemistry, representing an incremental improvement.
The paper tackles the challenge of slow training in temperature-conditional GFlowNets by proposing Logit-scaling GFlowNets, which accelerates training by scaling policy logits with a learned temperature function, achieving better generalization and mode discovery in biological and chemical tasks.
GFlowNets are probabilistic models that sequentially generate compositional structures through a stochastic policy. Among GFlowNets, temperature-conditional GFlowNets can introduce temperature-based controllability for exploration and exploitation. We propose \textit{Logit-scaling GFlowNets} (Logit-GFN), a novel architectural design that greatly accelerates the training of temperature-conditional GFlowNets. It is based on the idea that previously proposed approaches introduced numerical challenges in the deep network training, since different temperatures may give rise to very different gradient profiles as well as magnitudes of the policy's logits. We find that the challenge is greatly reduced if a learned function of the temperature is used to scale the policy's logits directly. Also, using Logit-GFN, GFlowNets can be improved by having better generalization capabilities in offline learning and mode discovery capabilities in online learning, which is empirically verified in various biological and chemical tasks. Our code is available at \url{https://github.com/dbsxodud-11/logit-gfn}