Knowledge Graph for Guidance and Explainability in Machine Learning

Project description

Machine learning (ML), especially deep learning, has achieved great success in AI and many industrial applications. However, successful deep learning models are still not fully transparent, and their predictions are often short of explanations, making them less trustworthy. Meanwhile, ML models often achieve great performance with standard datasets, but it still needs much domain knowledge support and guidance for data curation, (semi-)supervised learning, model selection, hyper parameter setting and so on, before they can be successfully deployed for practical usage. All these machine learning requirements for explainability and guidance are critical in many domains such as biomedicine and healthcare, where trustworthy decision making and expert guidance are highly required.

Knowledge graphs (KGs) are graph-structured knowledge resources including general purpose or commonsense facts, and conceptual knowledge about the application domain, ideally capturing that of a domain expert. Typically, this includes a hierarchy of concepts and also richer forms like axioms or rules. KGs have been widely used as the backend of AI and information systems, such as in Google search engine, Alexa and Siri. Many domains use ontologies (widely regarded as a kind of KG) to represent human knowledge with reasoning support, e.g., the SNOMED ontology in clinical settings. Integrating KGs with deep learning for neural-symbolic AI is a very promising direction with some pioneering studies conducted, showing the feasibility of using KGs for addressing many ML challenges such as the shortage of samples [1] and explanation [2].

The Information Management Group at the University of Manchester invites applications for PhD candidates in the area of Neural-symbolic AI with KGs. PhD projects in this area will explore how contemporary techniques in Machine Learning (such as Deep Neural Networks) and Symbolic AI (such as Knowledge Representation and Reasoning) can be used as a foundation to address the machine learning guidance and explainability challenges.

Examples of research challenges include: 1) how to construct a customized and robust KG for a machine learning task; 2) how to interpret a neural network model and justify its prediction for human beings; 3) how to inject domain knowledge into a machine learning for efficient learning with no or only a limited number of samples; 4) how to transfer models for a new task or context with domain knowledge guidance.

Applicants are expected to have:

1. An excellent undergraduate degree in Computer Science or Mathematics (or related discipline), and preferably, a relevant M.Sc. degree.
2. Confidence and independence in programming complex systems in Java or Python.
3. Previous academic or industry experience in at least one of the relevant topics such as Machine Learning, Knowledge Representation and Reasoning, Knowledge Graph, Semantic Web.
4. Excellent report writing and presentation skills.

Please note that applicants must additionally satisfy the standard requirements for postgraduate studies at the University of Manchester, such as a first-class or high upper-second class (or an equivalent international qualification) and English language qualifications, as stated in the PGR guidelines.

Qualified applicants are strongly encouraged to informally contact Jiaoyan Chen (jiaoyan.chen@manchester.ac.uk) or Uli Sattler (uli.sattler@manchester.ac.uk) to discuss the application prior to applying.

[1] Chen, J., etc. (2021) Knowledge-aware Zero-Shot Learning: Survey and Perspective. IJCAI Survey Track.
[2] Tiddi, I., & Schlobach, S. (2022). Knowledge graphs as tools for explainable machine learning: A survey. Artificial Intelligence, 302, 103627.