AnalysisGroupD

From canSAS
Revision as of 15:38, 14 June 2017 by Tim.snow (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

NOTES FOR GROUP D

Tim Snow Leading


Modelling 2D data

  • Everyone agreed that modelling 2D data has the potential to deliver more information than analysing 1D data, however, routinely analysing 2D data is not currently done
  • Fitting of 2D data requires careful thought if down sampling is required as features could be missed, conversely fitting a Pilatus 2M image will take a long time


Simulations and evolving algorithms

  • Projects coupling simulation with data fitting exist, although are typically in the earlier stages of development (i.e. SASSIE)
  • The next generation of algorithms are anticipated to revolve around either optimised simulation fitting or machine learning
  • Both approaches require a computing resource beyond a single desktop/laptop
  • Collaborations with institution HPC or national HPC are likely required
  • Additionally, links to commercial HPC/machine learning could be used for data fitting

Automation

  • Automation of data acquisition requires care as many variables change dynamically
    • Automatic alignment of some sample environments is challenging to human operators, let alone algorithms
    • Deducing when 'good' data has been obtained is tricky, especially if the sample is liable to beam damage
    • Determining an experimental endpoint could prove tricky too
  • Automation of static scans is a good starting point
  • Automatic data reduction should follow on from DAQ, however, fitting will likely require human intervention
  • Automatic creation and archival of sample meta-data would prove highly useful for machine learning


Collaborative funding strategies

  • Scientific funding is, as ever, extremely tight (however, also under increasingly close scrutiny as well...)
  • Combining efforts and collaborating is likely to give governments / funding bodies good reason to fund projects as facilities / institutions can point towards shared (i.e. free(ish)) resources and get 'more bang for your buck'
  • Finding common areas would be a positive step
    • Data analysis
    • Data fitting
    • Identification of data quality
    • Ways to automate different SAS measurements/techniques
    • Code development
    • Code maintenance/archiving
    • Experimental meta-data
    • Computing resource sharing (i.e. HPC)
  • Events, such as canSAS, a good way to start such links
  • Attending conferences either nationally or internationally a good way to form links with institutions, companies or funding bodies as well as other researchers