> ML_LITERATURE // SONG-2020-SCORE-BASED-GENERATIVE-MODELING-THROUGH-STOCHASTIC-DIFFERENTIAL-EQUATIONS_v1.0
Score-Based Generative Modeling through Stochastic Differential Equations
Yang Song, Jascha Sohl-Dickstein, Diederik P. Kingma, Abhishek Kumar, Stefano Ermon, Ben Poole · International Conference on Learning Representations (ICLR) (2020)
foundational2020foundationalthirdPartyReproduced
Principal Contribution
Unified score-based models and diffusion models under the framework of continuous-time stochastic differential equations (SDEs) and ODEs.
Operational Relevance
ICLR Outstanding Paper establishing Probability Flow ODEs for deterministic sampling, exact log-likelihood computation, and continuous time diffusion.
Assumptions
- Data perturbation across infinite noise scales converges to continuous forward diffusion governed by Itô SDEs
Limitations
- Predictor-corrector sampling can be computationally expensive; requires careful numerical SDE discretization
Connected Algorithms, Architectures & Tools
Related Algorithms:
Related Architectures:
Implementing Libraries: