Monday 27 July 2026 14:00¶

Flower AI: generally working, but what next?

Ways of working: for discussion

• Collab notes, then publish?
  • Write in Markdown instead of going via Google doc?
  • Hack.md or Framapad
  • Push to readthedocs or GitHub pages?

Definition of Done: Think about what we need to achieve, is this a single issue (one person) or is it bigger?

• bigger issues: break into sub-issues, or make it clear it's more like an epic/milestone
• focus on the outputs

TETREs: bottom up exploration of how to setup federated analysis

Flower AI: Two definitions of done (DoD):

• Do we have an exemplar analysis that we could test on? DoD: can this dataset analysis can be analysed
  • Suggested by Luis: multimodal US dataset: https://www.cancer.gov/ccg/research/genome-sequencing/tcga
  • Do our organisations each have some unique (but non-sensitive) data that we could run a distributed analysis on?
  • Could use CIFAR-10, split across sites, but need to make sure data is split in advance, then ingested into each site independently
    • Luis will look at splitting
  • NHS has artificial data https://digital.nhs.uk/services/artificial-data
• Is the deployment automatable? DoD: GitHub repo containing scripts/terraform/IaC etc