Concept demo — synthetic data — not an official site
AlzInsights Community (concept)

Learn

Cohorts built from what the pool needs

We looked at the gaps across all applications and built six-session cohorts around them, led by peers.

Privacy-preserving & federated methods

Few teams in the pool use federated learning, though many need it.

Peer leads

  1. Session 1

    Threat models for health data

  2. Session 2

    Federated averaging in practice

  3. Session 3

    Differential privacy budgets

  4. Session 4

    Secure aggregation tooling

  5. Session 5

    Auditing a federated run

  6. Session 6

    Show & tell

From prototype to validated system

Most teams are at prototype stage; validation is the common next step.

Peer leads

  1. Session 1

    What counts as validation

  2. Session 2

    External cohorts

  3. Session 3

    Pre-registration

  4. Session 4

    Calibration & drift

  5. Session 5

    Working with clinicians

  6. Session 6

    Write-up clinic

Agent evaluation on real cohorts

Agent-based approaches are popular; shared evaluation is rare.

Peer leads

  1. Session 1

    Defining agent tasks

  2. Session 2

    Ground truth on messy data

  3. Session 3

    Failure taxonomies

  4. Session 4

    Reproducible runs

  5. Session 5

    Human review loops

  6. Session 6

    Shared benchmark sprint

Only what you choose to share is visible. Contact details are never shown; requests go through the platform.

Replication challenges

Small, well-scoped projects that strengthen the evidence base. Open to every team.

Replicate a conversion-prediction result on a second cohort

ADNI → AIBL4–6 weeks

Re-run a biomarker panel with an open pipeline

ROSMAP3 weeks

Reproduce an agent's analysis plan end-to-end

AD Workbench2 weeks