Playgrounds › Noise Schedules › Laplace Schedule

Laplace Schedule

Originating Paper

Improved Noise Schedule for Diffusion Training

🧑‍🎓 Tiankai Hang, Shuyang Gu

📍 arXiv 2024

📄 34 citations

Why this schedule?

The Laplace schedule uses the CDF of the Laplace distribution to shape the noise trajectory. It concentrates the majority of noise injection around the midpoint of the diffusion process, meaning that relatively little noise is added at the very beginning and end, while mid-range timesteps see the sharpest transitions. Because the Laplace distribution is heavy-tailed, the schedule still provides meaningful noise variation across the full range of tt. This mid-concentrated approach can be computationally efficient, offering improved performance under resource constraints.

Formula

f(t; μ,b)=12bexp⁡ ⁣(−∣t−μ∣b)f(t;\,\mu,b) = \frac{1}{2b}\exp\!\left(-\frac{|t - \mu|}{b}\right)
  • •bb = scale parameter controlling the width of the distribution
  • •tt = current timestep
normalized values0.000.250.500.751.000.000.250.500.751.00
tt
αˉt\bar\alpha_t1.0000αt\alpha_t1.0000βt\beta_t0.0000SNR\mathrm{SNR}∞

βt\beta_t across all schedules

0.000.250.500.751.000.000.250.500.751.00
LinearCosineSigmoidExponentialCauchyLaplaceSquare RootLogistic

αˉt\bar\alpha_t across all schedules

0.000.250.500.751.000.000.250.500.751.00
LinearCosineSigmoidExponentialCauchyLaplaceSquare RootLogistic