Nested sampling from equivariant Graph Neural Networks

主要研究重点是准确理解为什么某些方法，如神经网络，适合于预测材料科学和分子建模领域的数据，结构变化如何影响这种能力，以及这种理解如何转化为更好的算法。这需要探索神经网络的参数空间。作为其中的一部分，将推进采样算法，探索分子和材料的构型空间，从而实现从第一原理预测属性。

The main research focus is to understand precisely why certain methods, such as neural networks, are suited to the task of predicting data in the fields of materials science and molecular modelling, how architectural changes influence this ability and how this understanding can be translated into even better algorithms. This entails the exploration of the parameter space of the neural networks. As part of this, sampling algorithms will be advanced which explore the configuration space of molecules and materials, enabling property prediction from first principles.