Ben studies the design, analysis and implementation of algorithms for time-dependent phenomena and modelling for problems in engineering and the sciences. His previous works has helped to establish the foundations of molecular simulation, providing efficient deterministic and stochastic numerical methods for an exploding field of application. A recent line of research has focused on stochastic algorithms for Bayesian inference from data and has demonstrated the potential for crossover of work from molecular science to data analytics (in this case molecular dynamics temperature controls were used to stabilise sampling-based parameterisation schemes).
Data science disrupts the traditional mathematical model by replacing physical law with empirical law, through the need to incorporate inference based on massive data sets or streams, and by destroying smooth structures (ODE/PDE solutions) that underpin numerical analysis. Data science is grounded in statistics and optimization, but to be effective in the engineering setting, data analysis methods such as Bayesian inference must be merged with models built up over many centuries on a foundation of physical law (e.g. quantum mechanics and thermodynamics) and be compatible with dynamical principles and geometric (or topological) constraints. During Ben's fellowship he plans to explore the interplay between "naive" data science approaches and models informed by physical law and mathematical structure. Access to The Alan Turing Institute is already providing him with connections to the statistics, operational research and machine learning communities. He looks forward to expanding these connections in the coming years.