I'm Xue Li.

I am studying Psychology part-time with a scholarship from the School of Informatics, the University of Edinburgh. Mental health plays such an important role in human's daily life. Better mental health support will hugely improve lots of people's well-being. However, support from human experts is limited compared with the need in society. Thus, I hope to apply AI technologies to assist Psychological experts for better mental health support. To make a better combination of AI with Psychology, I decided to take this MSc course to learn the knowledge from the field of Psychology.

A domain-independent algorithm was developed for repairing faulty Datalog-like theories by combining three existing techniques: abduction, belief revision and conceptual change (ABC). Based on devised mathematical models, the ABC system repairs faulty theories with better results than the individual techniques it combines because of its various operations: 1) add/delete axioms; 2) add/delete preconditions from rules; 3) change the language of the theory. Furthermore, ABC has wide applications, e.g. modelling game theory by abductively explaining given observations; root-cause analysis in system maintenance, and it has the potential to be extended to data in other formats.

I studied courses including Signoal Detection and Processing; Rada Systems, ect. My dissertation was about the development of a software system with UI, which forecasts the power range of high- frequency surface wave radar (HFSWR). Based on the mechanism of HFSWR detection, simulation models were established for the surface wave propagation, target echo, clutter jamming (environmental noise and sea clutter) and the signal-noise ratio of HFSWR. It was implemented in C/C# and Fortran languages. As a result, the user interface software predicts the scope of HFSWR.

I learnt foundation courses including Physics; Advanced Mathmatics and electrical engineering, ect. My dissertation was about the development of a face recognition algorithm based on machine learning was implemented based on machine learning algorithms, Discrete Cosine Transformation and Principal Component Analysis, in MATLAB.

Conducting Research and supervising students based on projects including 1) Extending the ABC automated repair system to address system failures based on the Knowledge Graphs; 2) Applying the ABC automated theory repairing system to legislation revision for autonomous vehicles; 3) Applying large lanugage models to detecting misinformation in counterfactual conditional claims.

Worked as an intern researcher on 1) aligning different knowledge graphs; 2) Analysing peer products to seek for new features; 3) Communicating with with other teams w.r.t. project development.

Developed scripts which generate Chinese versions of theorem certificates, as well as essential web maintenance.

My main work included software development and project management.

I supervised two Research Assistants and we conducted case studies on the application of the ABC repair system, an AI theory repair tool, to assist in constructing laws for autonomous vehicles (AVs) and adapting the system design of AVs to respect corresponding laws based on AVs’ simulations.

By integrating approaches from different subfields of computer science, namely, computational logic, deep learning, natural language processing and knowledge graphs (KGs), I am developing a detector to debunk misinformation on social media. Particularly, I focus on claims that involve the logic of causality. (e.g. “this would never have happened if…”).

I applied the ABC repair system to the root-cause analysis of system failures, which detects and adds missing information first and then suggests solutions to repair root causes. As the maintenance of a large system is expensive, our work contributes to making the task interactive between the ABC system and domain experts. Our team also worked on automatic KG construction with probabilities.

My second supervisors were Ewen Maclean (2016); Alan Smaill (2017-2019) and Eugene Philalithis (2019-2020). In this project, a domain-independent algorithm was developed for repairing faulty Datalog-like theories by combining three existing techniques: abduction, belief revision and conceptual change (ABC). Based on devised mathematical models, the ABC system repairs faulty theories with better results than the individual techniques it combines because of its various operations: 1) add/delete axioms; 2) add/delete preconditions from rules; 3) change the language of the theory. Furthermore, ABC has wide applications, e.g. modelling game theory by abductively explaining given observations; root-cause analysis in system maintenance, and it has the potential to be extended to data in other formats.