ALD-posterior
A sampling strategy in ALD that maintains dependencies between adjacent states by accumulating conditional variance and applying Gaussian bridge updates, more closely resembling the Markov chain structure of traditional diffusion.
Timeline (last 90 days)
The ALD paper studies two sampling processes, ALD-renoise and ALD-posterior, which share the same trained noise predictor and are distinguished only by different scalar inputs.
ALD-posterior maintains the dependency relationship between adjacent states through accumulated conditional variance and Gaussian bridge updates, and is closer to the Markov chain structure of traditional diffusion sampling.
All Facts (2)
The ALD paper studies two sampling processes, ALD-renoise and ALD-posterior, which share the same trained noise predictor and are distinguished only by different scalar inputs.
50%UnverifiedALD-posterior maintains the dependency relationship between adjacent states through accumulated conditional variance and Gaussian bridge updates, and is closer to the Markov chain structure of traditional diffusion sampling.
50%