12 June: Greg McInerny (CIM)

BACKFILLZ and WAYS – Addressing longstanding visualisation problems at the site of production.

 

I will introduce the BACKFILLZ project that James Tripp and I have been working on. Here we address some issues in MCMC visualisation (MCMC, Markov Chain Monte Carlo is an optimisation/parametrisation algorithm that is widely used in data science). The problem with Monte Carlo is that there is no definitive way to know if the algorithm has been run for long enough and if you have enough samples. Our goal is to articulate concepts such as ‘mixing’ and ‘ergodicity’ in the visualisation and increase the information available to the user. This is difficult as MCMC chains contains so many samples, and we might have multiple chains and many parameters.

In the second half I will introduce the new project that the Alan Turing Institute have agreed to fund, which relates to the precursor visualisation concept that I have mentioned to some people previously. Here we are addressing the issue that visualisation software does not necessarily produce visualisations. In the WAYS project James and I will work with people in the Social Justice and Digital Twins: Data Centric Engineering themes at Turing, as well as conducting research on how people visualise data using a new spyware called GRAPHO. Its an exciting new project that will last 18 months, and hopefully start in September.