Computational Statistics and Machine Learning

This theme is concerned with advancing the theory, methodology, algorithms and applications to modern, computationally intensive, approaches for statistical inference.

We encompasses topics from machine learning methods for adaptive systems and flexible modelling, to optimisation and simulation based approaches to solve intractable computational problems of relevance to statistics, such as variational or Monte Carlo methods and simulation-based inference.

The need for advanced algorithms is a reality in any modern day application of statistics, be it in finance, social science, natural sciences or beyond. The development of simulation and optimisation algorithms, as well as the analysis of their properties in the context of statistical inference, is a core mission of our group. Our research has advanced the applicability of models that imply complex posterior distributions in Bayesian inference, or more generally those that involve sophisticated generative models required to faithfully capture the structure of real data. 

Machine learning originated from research in artificial intelligence, where historically the focus has been on data-driven approaches for the development of autonomous systems. However, it has grown to affect science and society in many high-stakes applications that require careful statistical diligence and a deep understanding of the specific needs raised by real-world data. Our group advances machine learning methodology by embracing the adaptability and generality of AI-inspired research with the expectations of stakeholders who require robustness and uncertainty quantification in their applications.

Theme Members

• Thomas Barlett
• Francois-Xavier Briol
• Purvasha Chakravarti
• Petros Dellaportas
• Thomas Honnor
• Jeremias Knoblauch
• Brieuc Lehmann
• Samuel Livingstone
• Ioanna Manolopoulou
• Giampiero Marra
• Yvo Pokern
• Javier Rubio
• Ricardo Silva (Theme Lead)
• Jinghao Xue
• Jialin Yu

Research Groups

• Fundamentals of Statistical Machine Learning. The focus of this research group is on the intersection of statistical inference and machine learning methodology and theory. 
• Data, Environments, and Learners: Theory and Algorithms. The group's interests revolve around machine learning theory and practice, with a particular interest in learning and certification strategies that are reliable and make efficient use of the available data.

Recent and Upcoming Events

• London Meeting on Computational Statistics, UCL, June 2024
• Computational Statistics Theme Day 2024, UCL, April 2024
• Distance-Based Methods in Machine Learning Workshop, UCL, June 2023

Current and Recent Externally Funded Projects

• Transfer Learning for Monte Carlo Methods, EPSRC New Investigator Award EP/Y022300/1. June 2024 - June 2027. PI: Briol.  
• Robust Foundations for Bayesian Inference, EPSRC EP/Y011805/1. June 2024 - June 2025. PI: Briol, co-I: Knoblauch.
• CHAI - Causality in Healthcare AI with Real Data, EPSRC AI Hub EP/Y028856/1, February 2024 - January 2029, Co-Is: Silva and Karla Diaz-Ordaz
• The Causal Continuum - Transforming Modelling and Computation in Causal Inference, EPSRC Open Fellowship EP/W024330/1, October 2022 - September 2027, PI: Silva
• Robust and Scalable Markov Chain Monte Carlo for Heterogeneous Models, EPSRC EP/V055380/1. July 2022 - October 2024. PI: Livingstone
• Geometry and Bayesian statistics to reconstruct protein radical structures from ENDOR spectroscopy, Collaborative Research Center 1456, Göttingen, Fellow: Pokern