Phil Marshall Meeting 2017/06/29

Validating Bigmali At Scale Series:

• Validating Bigmali At Scale
• Validating Simpler Model
• Validating Likelihood Variance And Single Likelihood Weight Distribution
• Validating Simpler Model With Many Samples
• Validating Full Model With Many Samples

Mass Mapping:

• Mass Maps From Mass Luminosity

Information Analysis:

• Information Metrics And Bounds

Miscellaneous:

• Now a regular Sherlock consumer. Looking forward to scaling up lots of stuff moving forward.
• Already seeing return on investment from Ryden (~70pgs so far) and Schneider (~40pgs so far, big pages)
• Read H0LiCOW 5, looking forward to H0LiCOW 8
• Lightly skimmed Reddick's Thesis
• Noticed Probabilistic Cosmological Mass Mapping From Weak Lensing Shear, looks awesome.
• Saw "Phil Marshall was elected as the new spokesperson of the LSST Dark Energy Science Collaboration." - KIPAC Newsletter. Congrats.

Meeting Summary and Next Steps

• Read up on Gellman effective sample size (ESS).
  • Phil sometimes uses 95% of weight as a proxy.
  • Summarize for Phil.
• Wrapping Up Bigmali Mass-Luminosity Inference
  • Do pen and paper analysis of error accumulation/propagation in bigmali.
    • Test assumption: in low dimensions bigmali problems go away.
    • See if divide and conquer could be better.
    • Do analysis with Gaussians.
    • Model bias and uncertainty at each step
  • Final Report?
• Mass mapping notebook
  • Incorporate pangloss/massinference
  • Review Pangloss to see how its functionality connects.
  • Characterize scaling.
  • Characterize approximation - can we use normal distribution to approximate halo mass distributions?
    • Is the surface smooth? Then we could use emulator.
• Ask Phil for the following resources:
  • Magic Beans Tarball
  • Draft of H0LiCOW 8
• Read Mcully + Wong (Weak Lensing) and Schnieder (probabilsitic mass mapping ...)
• Other topics:
  • Given David's background and ICME training, fast, large scale linear algebra could be an area in Weak Lensing where we could offer a unique value proposition to the community.