Decision Tree Induction from Small Language Models for Black-Box LLM Adaptation

About the Project

Project Description

Flagship black-box large language models (LLMs) such as GPT-5 and Gemini-3 perform well across a wide range of tasks, including those that require reasoning or generating explanations. However, they can be unreliable on specialised, domain-specific text classification tasks (e.g., sentiment analysis, hate-speech detection), particularly when in-domain data is limited or decision boundaries are subtle. Fine-tuning is not always possible due to closed-model restrictions, safety policies, or cost, which has motivated research into methods that can adapt such models without updating their weights. In contrast, smaller language models (SLMs) can be efficiently fine-tuned on in-domain datasets and often achieve substantially higher classification accuracy as a result, but their decisions are harder to interpret, and they typically produce less reliable explanations. This limits their use in settings that require trustworthy AI.

This project explores a neuro-symbolic approach to bridge this gap by extracting decision trees of concepts from specialised SLMs and using these trees to adapt black-box LLMs.

Aim: Develop methods to induce decision trees of high-level concepts from fine-tuned SLMs, and use these trees as a structured representation of the decision boundary learned from training data. The goal is to assist black-box LLMs in making classification decisions in specific domains and to enable them to generate faithful explanations.

Methods: The student will: (1) fine-tune SLMs for domain-specific tasks (e.g., text classification); (2) automatically induce high-level concepts from SLM embeddings and label/refine these concepts using LLM-assisted labelling; (3) organise the concepts into a decision tree that reflects the SLM’s learned decision boundary; (4) develop inference-time mechanisms to allow a black-box LLM to use the induced tree to support domain-grounded classification decisions and faithful explanations; and (5) evaluate task performance and, critically, explanation faithfulness and robustness using counterfactual testing, concept interventions, and human evaluation of explanation quality.

Indicative deliverables:

• Algorithms for inducing concept-based decision trees from fine-tuned SLMs
• A prototype system for LLM decision-making and explanations grounded in domain-specific training data
• An evaluation framework for faithfulness, robustness, and user trust
• Open-source code and peer-reviewed publications

The successful candidate will join Cardiff University’s vibrant NLP and KR community and be supervised by Dr Nitesh Kumar and Prof Steven Schockaert.

Keywords: interpretable ML; concept discovery; decision tree induction; NLP; LLM grounding; neuro-symbolic AI; trustworthy AI; LLM adaptation.

How to Apply

This project is accepting applications all year round, for self-funded candidates.

Mode of Study: Full-time or part-time

Please submit your application via Computer Science and Informatics - Study - Cardiff University

In the funding field of your application, indicate “I am applying for a self-funded PhD in Computer Science and Informatics”, and specify the project title and supervisors of this project in the text box provided.

Academic criteria: A 2:1 Honours undergraduate degree or a master's degree, in computing or a related subject. Applicants with appropriate professional experience are also considered. Degree-level mathematics (or equivalent) is required for research in some project areas.

Applicants must demonstrate English language proficiency. Students who do not have English as a first language must prove this by obtaining an IELTS score of at least 6.5 overall, with a minimum of 6.0 in each skills component. A full list of accepted qualifications is available here: https://www.cardiff.ac.uk/study/international/english-language-requirements/postgraduate

If you are interested, please contact Dr Nitesh Kumar (kumarn8@cardiff.ac.uk) sending your CV in the first instance. The application process requires you to develop an individual research proposal jointly with the supervision team, which builds on the information provided in this advert.

Once you have developed the proposal with support from the supervisors, please submit your application following the instructions provided below.

Please submit your application via Computer Science and Informatics - Study - Cardiff University

In order to be considered candidates must submit the following information:

• In the ‘Research Proposal’ section of the application enter the name of the project you are applying to and upload your Individual research proposal. Your research proposal should not exceed 2000 words, including references and bibliography.
• A personal statement (as part of the university application form, or as a separate attachment, if you prefer).
• A CV. Guidance on CVs for a PhD position can be found on the FindAPhD website.
• Qualification certificates and Transcripts - original and English translation, if applicable.
• References x 2 which should be academic references. Please note you need to provide the reference documents as part of your application.
• Proof of English language (if applicable).

Interview– If the application meets all of the entrance requirements listed above, you will be invited to an interview.

Funding Notes

This project is offered for self-funded students only, or those with their own sponsorship or scholarship award. Where applicable, candidates will be required to cover the cost of a student visa, the healthcare surcharge, and any other costs of moving to the UK to study. These costs will not be covered by the School of Computer Science and Informatics.