VARD fine-tunes diffusion models by backpropagating through a learned value function that assigns dense, differentiable reward estimates to every intermediate denoising step, with KL regularization keeping the model near the pretrained weights.
Atom level enzyme active site scaffolding using rfdiffusion2
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
VARD: Efficient and Dense Fine-Tuning for Diffusion Models with Value-based RL
VARD fine-tunes diffusion models by backpropagating through a learned value function that assigns dense, differentiable reward estimates to every intermediate denoising step, with KL regularization keeping the model near the pretrained weights.