Neural spline flows, conditional flow matching, and diffusion models each win on different molecular benchmarks: flows handle asymmetric low-dimensional distributions, flow matching handles high-dimensional simple data, and diffusion handles complex low-dimensional data.
Krishna , author J
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
A survey of probabilistic generative frameworks for molecular simulations
Neural spline flows, conditional flow matching, and diffusion models each win on different molecular benchmarks: flows handle asymmetric low-dimensional distributions, flow matching handles high-dimensional simple data, and diffusion handles complex low-dimensional data.