Conditional Diffusion with Less Explicit Guidance via Model Predictive Control
This work addresses a specific challenge in conditional diffusion models for researchers and practitioners, presenting an incremental improvement in efficiency.
The paper tackles the problem of conditional sampling with limited explicit guidance by using a model predictive control (MPC)-like approach to approximate guidance, showing that MPC-approximated guides achieve high cosine similarity to real guides and improve generative quality when explicit guidance is restricted to five time steps.
How much explicit guidance is necessary for conditional diffusion? We consider the problem of conditional sampling using an unconditional diffusion model and limited explicit guidance (e.g., a noised classifier, or a conditional diffusion model) that is restricted to a small number of time steps. We explore a model predictive control (MPC)-like approach to approximate guidance by simulating unconditional diffusion forward, and backpropagating explicit guidance feedback. MPC-approximated guides have high cosine similarity to real guides, even over large simulation distances. Adding MPC steps improves generative quality when explicit guidance is limited to five time steps.