2019 · Generative models
Generative Modeling by Estimating Gradients of the Data Distribution
Score-based generation learns directions that move noisy samples toward the distribution of real data.
Paper & contextYang Song
Generative modelsOfficial profile
Learning score functions to guide generative sampling.
Song studies score-based generative modeling: learning directions that move noisy samples toward data. This selection follows that idea through stochastic differential equations, inverse problems, and consistency models designed to produce samples in fewer steps.
Selected work
10 papers
2020 · Generative models
Score-Based Generative Modeling through Stochastic Differential Equations
This paper unifies score-based generative methods through a continuous process that adds noise and a learned process that reverses it.
Paper & context2023 · Language models
Consistency Models
Consistency models aim to generate data in one or a few steps by learning mappings that agree along a denoising trajectory.
Paper & context2023 · Generative models
Improved Techniques for Training Consistency Models
This paper improves consistency-model training so models can learn directly from data with stronger one- and two-step generation.
Paper & context2024 · Generative models
Simplifying, Stabilizing and Scaling Continuous-Time Consistency Models
Continuous-time consistency models are simplified and stabilized to support larger-scale training and fast image generation.
Paper & context2021 · Generative models
Solving Inverse Problems in Medical Imaging with Score-Based Generative Models
Score-based generative priors are used to reconstruct medical images from incomplete measurements.
Paper & context2021 · Generative models
Maximum Likelihood Training of Score-Based Diffusion Models
This paper connects weighted score-matching objectives with likelihood training for diffusion models.
Paper & context2020 · Generative models
Accelerating Feedforward Computation via Parallel Nonlinear Equation Solving
Feedforward computation is reframed as solving nonlinear equations so parts of an otherwise sequential calculation can run in parallel.
Paper & context2020 · Generative models
Improved Techniques for Training Score-Based Generative Models
This follow-up improves the stability and resolution of score-based image generation.
Paper & context2019 · Generative models
Sliced Score Matching: A Scalable Approach to Density and Score Estimation
Sliced score matching uses random projections to make score estimation more computationally tractable.
Paper & contextAn independent editorial profile. Inclusion does not imply Council membership or endorsement. Research is collaborative; coauthorship does not imply sole credit.
Selection & sources