Thema der Dissertation:
Efficient Generation of Molecular Boltzmann Distributions with Machine Learning Methods Thema der Disputation:
Deep Generative Modeling with Normalizing Flows
Efficient Generation of Molecular Boltzmann Distributions with Machine Learning Methods Thema der Disputation:
Deep Generative Modeling with Normalizing Flows
Abstract: In recent years, normalizing flows have emerged as a powerful class of generative machine learning models, with applications ranging from image and audio synthesis to problems in the natural sciences, such as generating samples from equilibrium distributions or designing de novo molecules.
The first talk introduces the theoretical foundations of normalizing flows, which model complex probability densities by applying invertible transformations, often parameterized by deep neural networks, to map simple prior distributions to more complex target distributions. The discussion will focus on two main formulations: coupling flows and continuous normalizing flows.
The second talk highlights applications of normalizing flows for sampling from equilibrium Boltzmann distributions. It covers two principal strategies: using flows to propose large timesteps in molecular simulations and generating independent equilibrium samples directly from the Boltzmann distribution.
The first talk introduces the theoretical foundations of normalizing flows, which model complex probability densities by applying invertible transformations, often parameterized by deep neural networks, to map simple prior distributions to more complex target distributions. The discussion will focus on two main formulations: coupling flows and continuous normalizing flows.
The second talk highlights applications of normalizing flows for sampling from equilibrium Boltzmann distributions. It covers two principal strategies: using flows to propose large timesteps in molecular simulations and generating independent equilibrium samples directly from the Boltzmann distribution.
Zeit & Ort
15.08.2025 | 14:00
Hörsaal A (1.3.14)
(Fachbereich Physik, Arnimallee 14, 14195 Berlin)
&
WebEx