Robert McGreevy, Rutherford Appleton Laboratory, reflects on how LINXS can support the use and DEVELOPMENT of neutron and X-ray based RESEARCH
Robert McGreevy, based at the Rutherford Appleton Laboratory in the UK, gave a guest lecture at LINXS in early December. He is a world-leading researcher in the area of neutron scattering and computer modelling, and has a special interest in the development and integration of data analysis and data management tools with experiments.
Robert McGreevy is based at the Rutherford Appleton Laboratory in the UK.
He can see many ways in which LINXS can support the use, and further development, of neutron and X-ray based research.
– LINXS is a great place for bringing people together: to give them the space to develop new ideas. Just to take an example: after my seminar, we had a very fruitful discussion on ways to improve data analysis with the help of AI, which I will now take further.
LINXS can be an arena for conversations
Providing an arena for conversations that generate new scientific progress is important, according to Robert McGreevy, because so many events today focus on short talks and presentations, and very little time is spent on proactive discussion on scientific issues. His own area of interest, software development and data integration, is one where there is a big need for more focused efforts.
– As I see it, we need to develop how we think about the use of data in relation to experiments. People think it is a straight line: you do an experiment, you run some software, you get an answer. But the advanced stuff is actually very complicated, and requires innovative modelling, and information needs to be pulled from different sources. If we want to do that well, and to innovate, we need to get people to use their imagination, and think about these things on a deeper level, he says.
He adds: – LINXS is not going to run big computers or big data sources, but LINXS can play an important role here; as a convenor of discussions, and as a facilitator to get these kinds of conversations going.
LINXS can support the ESS in engaging neutron users
Another opportunity he can see for LINXS is to support the ESS in engaging neutron users to get them ready to use the instruments once the ESS opens.
– Most researchers work on a one or two-year time-line as needed for their PhD or post-doc projects. This may mean that their focus is not yet on the science they could do at the ESS in five years time. However, it is crucial to engage longer-term thinking, to encourage and stimulate the type of scientific profile that will need to be in place by the time ESS opens, he says.
Ways that LINXS could facilitate this effort, according to Robert McGreevy, could be to host workshops, or work more closely with the ESS to develop an interesting seminar programme around neutron techniques and research frontiers - with seminars meaning a real discussion, not just a lecture.
He adds that LINXS should also continue its work of highlighting research and developments from different neutron and X-ray facilities.
– Research to generate new ideas on how to use neutron and X-ray science is ongoing at many places in Europe and elsewhere. As such, I think that it is very important that LINXS disseminates opportunities and discoveries across a range of facilities. In doing so, LINXS can become a strategic partner, not just for MAX IV and ESS, but for other facilities too.
Use real data from facilities to develop AI and data development
So how does he plan to take his idea of AI and data development further?
– Martin Stankovski at LINXS and I had an idea to collect real benchmark data sets at various facilities, and use those sets to develop the use of AI in data collection and analysis during experiments. Machine learning benchmarks datasets are often model data which do not contain any errors. That means your machine learning techniques might not work properly once they are tested in the real world.
– With real data, AI might become better at synthesizing data. But to do this properly we would also have able to develop full metadata sets, which fully describe the parameters of the experiment. This type of data is currently very rare.