Current and planned deblenders for LSST employ non-parametric models and analytic constraints to determine the most plausible blend configurations, but galaxies have complex shapes. A closed-form description of galaxy morphologies is elusive, which means that implicit descriptions (e.g. from neural networks) or more sophisticated interpolations (e.g. Gaussian processes) might help. Alternatively, we might think of not doing deblending at all. Moreover, there is a need for optimized de-blending approaches for variable sources in galaxies. This session is meant to explore ideas that could lead to radically different concepts of how to deal with blending.

Agenda

• Peter Melchior: Introduction and unsolved problems
• Valerio Roscani: ASTErIsM
• Sowmya Kamath: Detection on residuals
• Erin Sheldon: Tests of the Scarlet and Multi-object Fitting Deblenders for
  Shear Recovery
• Tony Tyson & Erfan Nourbakhsh: Don’t deblend at all, implications for DE metrics
• open discussion

Google doc for scribe notes (key findings, next steps, actionable items).