Menu Close

Dr. Michael Knight

Malvern Panalytical

Dr. Michael Knight is a senior data scientist based at the Bristol (UK) site of Malvern Panalytical. He specialises in the development of data analysis methods for a diverse range of scientific instruments including laser diffraction, calorimetry and near infra-red spectroscopy. He focuses on leveraging cloud and high-performance computing to deliver insight from data.

Michael joined Malvern Panalytical in 2019. He obtained his BSc in biochemistry with chemistry in 2006, and PhD in biophysics in 2009, both from the university of Southampton (UK). 

He held academic research positions at Ecole normale superieure de Lyon focussing on solid-state NMR of proteins (2010-2012) and the university of Bristol focussing on quantitative MRI (2013-2017). He moved to industry in 2017. 

Michael has authored over 40 peer-reviewed publications and is a patented inventor.

Presentation:

How could (and should) machine learning be integrated into particle sizing technologies

Particle size is rarely measured directly. It is usually derived as a model-dependent estimate from observations of some physical quantity that is sensitive to it – for example laser or x-ray scattering. In every particle sizing technology, somewhere or other, there is a model to convert observation into insight. Those models depend on assumptions, and on parameters that have been tuned from exposure to data, both of which leave open questions. The ubiquity of machine learning (ML) in modern science motivates us to ask to what extent it could (and should) play a role in particle sizing. This talk will present recent approaches to the incorporation of ML methods into the particle sizing process, at various stages from data acquisition to size estimation. It will show how the careful choice of methods can reduce the number of assumptions and requirement for custom configuration of particle sizing algorithms on a per-sample basis. It will consider cases where we can make particle sizing more transparent and robust. 

×