Training, fine-tuning, and AI supply-chain security
Establish trust before deployment.
How can AI systems retain security as training data, fine-tuning updates and supplied components change?
An AI system depends on data, base models, fine-tuning processes and software supplied by several sources. We explore how to assess those dependencies, trace changes through the development process and evaluate whether an update introduces unwanted behaviour before it reaches users.
Research areas
Study data and model integrity across collection, training and fine-tuning.
Examine poisoning, backdoor risks and unintended behaviour introduced by updates.
Investigate provenance and dependency tracking for models, datasets and supplied components.
Evaluate release checks and continuing assurance as AI dependencies change.
Publications & related resources
Projects, publications and research materials will be added here.
Contribute to this research.
Explore student research, internships and collaboration with the lab.