Scientific machine learning for information extraction from observation data

Machine learning has become an indispensable tool for extracting useful information from massive amounts of data, which makes it become an integral part of industries and research fields. However, traditional machine learning techniques often fail to fully integrate domain-specific knowledge and logical reasoning into the learning process. Information Extraction (IE) is a vital research area that focuses on generating structured information from natural language inputs. While many researchers have proposed deep learning approaches to address the IE task, these methods lack the ability to incorporate established logical relations as training constraints. To overcome these limitations, this project explores the emerging field of Scientific Machine Learning (SML) for IE from observation data. Specifically, we implement a transformer-style deep neural network that incorporates logical knowledge in the form of First-Order Logic (FOL) for joint training. Compared to traditional deep learning models, our logical model significantly improves the effectiveness of the IE task.Bachelor of Engineering (Information Engineering and Media