Por Haroldo V. Ribeiro (Universidade Estadual de Maringá, Brasil).
The use of machine learning methods is becoming increasingly important for the development of materials science and is now part of the agenda of physicists. These methods have already proven to be helpful in uncovering materials properties and simplifying experimental protocols; however, several areas have taken little to no advantage of these approaches. That is the case of research involving liquid crystals, which is somewhat surprising as these materials are often investigated via imaging techniques, and it is precisely with images that machine learning methods have achieved breakthroughs in recent years. This talk presents some recent works that have tried to reduce this gap by combining ordinal methods and machine learning algorithms to estimate liquid crystal properties directly from their optical textures. We show that ordinal methods (namely: complexity-entropy plane and ordinal networks) produce good representations for liquid crystal textures, allowing simple learning algorithms to create very precise maps between images and physical properties of these materials.
Transmissão em direto via Zoom.
