Deep Learning for Computer Vision and Applications in Materials Informatics

Yang, Zijiang

doi:https://doi.org/10.21985/n2-wt4v-jf75

Work

Deep Learning for Computer Vision and Applications in Materials Informatics

Public

Download PDF

Download All Files (.zip)

Deep learning is a new area of machine learning in artificial intelligence that consists of networks to learn representations from data in a supervised, semi-supervised and unsupervised manner. Deep learning has a relatively long history, but it does not gain great attention until big data and fast computational resources are available. Due to the flexible architecture and striking learning capability, deep learning has now become the state-of-the-art machine learning approach in various fields of computer science, such as computer vision and natural language processing. Particularly, the convolutional neural network is widely used in various research tasks in computer vision, such as image classification and segmentation, object detection and automatic driving. Due to its great success in computer science, deep learning has been also applied in other scientific fields, such as biology, chemistry. Although deep learning has become more and more popular, it is still an emerging area in materials science research. Materials informatics is a multidisciplinary science combining materials science, statistics and computer science that utilizes data-driven methods to explore the inherent nature of materials systems. Broadly speaking, there are two problems in materials informatics, which are forward modeling and inverse modeling. There are many challenges in both directions, such as complexity of microstructure, lack of big data for forward modeling and design of microstructural representation and optimization in high-dimensional materials space for inverse modeling. Thus, this thesis explores the theory and applications of deep learning in various computer vision problems of materials science in the domain of both forward and inverse modeling.

Creator