ML@CL seeks to advance the safe and reliable deployment of machine learning systems to tackle the real-world challenges.
This site summarises current activities from the students and researchers working with Carl Henrik Ek, Ferenc Huszar, Neil Lawrence and Jessica Montgomery. Our work addresses the full pipeline of AI system development, from data acquisition to system deployment, advancing these research ideas in a wider context that considers the availability, quality and ethics of data use and the ways in which machine learning can be deployed for public benefit. Areas of interest include:
Machine learning: The machine learning systems that have driven much recent progress in AI learn from data about us and the world in which they operate. As systems using machine learning become larger and more complex, and are deployed into uncontrolled environments, challenges in interpretation, reliability, accuracy and fairness arise. These demand fresh approaches to monitor system performance, embed explainability, and build system resilience. Together, these approaches point to a paradigm shift in computer science, revisiting assumptions and approaches to systems design, programming languages and systems security.
Data: The availability and quality of data influences the effectiveness of machine learning systems. There are opportunities now to develop machine learning methods that are better able to deal with messy or poor-quality data that systems often encounter in real-world environments. Data oriented programming, for example, offers a set of development methodologies to ensure system designers account for: what decisions are required, how they will be made, and how these interconnect within the system architecture. This creates the possibility of a more sophisticated form of auto ML, where full redployments of models are considered while analyzing the information dynamics of a complex automated decision-making system.
Policy: Policy plays a crucial role in influencing where, how and for whose benefit machine learning systems are developed and deployed. Safe and reliable deployment of machine learning systems requires policy frameworks that embed trustworthy data governance, connecting our aspiration to share data for personal or public benefit with our desire to protect individual rights; that promote the use of machine learning in areas where it has potential to improve public services and wellbeing; and that account for the wider implications of technological change on individuals and communities.
Science: By analysing complex datasets and uncovering previously unknown patterns, machine learning has the potential to accelerate scientific discovery across the sciences – from healthcare to climate science, fundamental physics to conservation, and more. The use of AI for scientific discovery offers opportunities to create exciting new research agendas that both advance data science methods and generate new insights for research. Realising this potential requires action to equip researchers from across disciplines with the skills they need to use machine learning in their work, and to build a community of practice at the interface of data science and other disciplines.