Temporal Alignment Guidance: On-Manifold Sampling in Diffusion Models
This addresses a fundamental limitation in diffusion models for generative AI applications, though it appears incremental as it builds on existing guidance frameworks.
The paper tackles the problem of error accumulation and off-manifold sampling in diffusion models when applying arbitrary guidance, which breaks sample fidelity. The proposed Temporal Alignment Guidance (TAG) method consistently produces samples aligned with the desired manifold at each timestep, leading to significant improvements in generation quality across various downstream tasks.
Diffusion models have achieved remarkable success as generative models. However, even a well-trained model can accumulate errors throughout the generation process. These errors become particularly problematic when arbitrary guidance is applied to steer samples toward desired properties, which often breaks sample fidelity. In this paper, we propose a general solution to address the off-manifold phenomenon observed in diffusion models. Our approach leverages a time predictor to estimate deviations from the desired data manifold at each timestep, identifying that a larger time gap is associated with reduced generation quality. We then design a novel guidance mechanism, `Temporal Alignment Guidance' (TAG), attracting the samples back to the desired manifold at every timestep during generation. Through extensive experiments, we demonstrate that TAG consistently produces samples closely aligned with the desired manifold at each timestep, leading to significant improvements in generation quality across various downstream tasks.