Single-sample unmixing with machine learning: a rock magnetic frontier

磁性造岩矿物可以记录地球磁场和气候变化的重要信息。在岩石磁学中，我们试图量化地质材料中的磁性变化。现有的量化方法是有限的，并提供大量的材料中的所有磁性颗粒的特性，而不是诊断信息有关的个别矿物成分。该项目旨在开发一个机器学习框架，以“解混”和量化单个天然样品中的每种磁性矿物成分，并将开启岩石磁性定量的新时代。预计它将通过在物理学，材料科学和工业中实现磁性表征而产生超越地球科学的影响。

Magnetic rock-forming minerals can record important information about Earth’s magnetic field and climatic changes. In rock magnetism, we seek to quantify magnetic property variations in geological materials. Existing quantification methods are limited and provide bulk characterisation of all magnetic particles in a material rather than diagnostic information concerning individual mineral components. This Project aims to develop a machine-learning framework to “unmix” and quantify each magnetic mineral component in single natural samples, and will unlock a new quantitative era in rock magnetism. It is expected to have impact beyond Earth science by enabling magnetic characterisation in physics, materials science, and industry.