22nd July 10:35 – 10:50
Speaker: Professor Georgina Cosma
Abstract: This presentation explores secure artificial intelligence (AI) techniques for government and sensitive information systems. Traditional AI approaches rely on external APIs and unverified training data, creating security vulnerabilities. Key techniques include purpose-built Retrieval Augmented Generation (RAG) systems using only curated, internal data sources with zero external dependencies. These secure techniques address the considerable challenge of processing vast government archives through semantic search capabilities that understand context beyond simple keyword matching. The presentation covers advanced approaches including agentic RAG for autonomous decision-making and multi-modal processing techniques. We also examine the complex challenge of “unlearning” – removing specific information from AI systems. These techniques demonstrate how AI can be implemented safely within sensitive environments whilst maintaining transparency and accountability.
Bio: Georgina Cosma is a Professor of AI and Data Science at Loughborough University, UK. She holds a PhD in Computer Science from the University of Warwick. Her research focuses on neural information retrieval, natural language processing, and machine learning and unlearning for text and multi-modal data. She leads the Neural Information Processing, Retrieval & Modelling research group and has secured funding from UKRI, NIHR, the RAF, the Health Foundation, the Leverhulme Trust, and the European Commission. Her work tackles challenges in healthcare, public sector AI, and responsible and ethical AI.