Thoughts on BigData/InferenceSummerSchool/KTS

Two basic issues:
1)  where does the disconnect between the astronomers and the ML/CS folks actually occur?
2)  are the any women of any seniority on the astro/data-mining world

On 1)
HWR's suspicion is that astronomers almost always deal with data that are noisy,
 and heteroscedastic at that.
Therefore, astronomers know how to write down likelihood, go Bayes, but: and then??
CS/ML folks have amazing tools to classify, but these tools almost all fail ungracefully in the
"very noisy" regime. I.e. are not good at simply ignoring differences in object labels that can sensibly
attributed to only noise, not to any inherent difference of the objects.

What's a good definition of noisy here:
Let's presume any object has many data points (e.g. pixels in the spectrum of a star);
there is the regime where the data variance at any one pixel due to noise is comparable
to the (to be classified) ensemble variance (noiseless) in that 'pixel'. What are good ML tools
in that regime.

What to do about it
a) try to spell this out clearly, and ask Hogg, DFM, ZI, BernhardSchöllkopf etc. for insights?
b) initiate some culture-gap bridging exercise with the HITS group, to see how useful that is..


On 2)
The Tel Aviv Big Data conference (Dec. 15), the search for lecturers for an IMPRS summer school,
showed the paucity of "obvious" female scientists to serve as lecturers or tutors.
What to do?
-- find out whether this is just a consequence of HWR's ignorance? Ask Hogg, Schöllkopf, Ivezic, Bailer-Jones, DFM for names...
-- initiate discussion with Reutter to see whether there might be any interest in initiating
   an "award", amounting to a 3-year fellowship, for women in data science.
  To be taken as a post-doc, or a repeat summer fellowship.

Tel Aviv notes

Post Tel Aviv

-- M-stars WD's with Dani

-- Cannon residual into Nikos Gianniotis' machinery

-- write-up for Chao Liu's PD.

PS1 Cepheids revisited

After doing some more thinking, learning a lot from Laura already, here's my current thinking about Cepheids:

1) in the context of Galactic disk evolution, Cepheids could should play the role of the ideal 'present-day', young star diagnostic.
They can answer: where and with what (detailed!!) abundances are stars born now.
2) With them, we can do the dynamics of "the other side of the galaxy" .

By now, quite a number of things are in the AS4 white paper by Byrd

Nano-Jasmin: ESA to participate in Japan's mini-Hipparchos?

Japan's National Observatory is planning to launch a 35kg astrometric satellite, dubbed Nano-Jasmin, to do astrometry at the very bright end. The idea might be: if ESA provides the launch, the the "Gaia team" will get access to the data. Ask Coryn what he knows about this...

LSST in Germany

After conversation with Steve Kahn, Michael Strauss (& at end -- Joe Mohr)

Basic plan:

get a DPG (not the whole community) going, which involves the usual suspects
and enough university participation that a DFG Schwerpunkt, as suggested by Krückeberg,
seems feasible..

DFG Schwerpunkt, deadline Oct 15; decision by March; Rundgespräch recommended


HWR's action to coordinate within the MPG:

-draft e-mail with the basic options:
Who want's to commit how many "individual slots"
Should we cast this is terms of a joint MPG participation group?
Should we approach the MPG Präsidium for central funds?

What do spectra tell us about age (that is not encased in [Fe/H], [a/Fe])?

just a note to be brought up with Melissa, to commit to memory until she's back from Michigan:

Basis: Cannon trained with 5 labels, with ages from Kepler; with the Cannon, we get "age" recovered as a parameter with +-2 Gyrs. BUT: to which extent is the age just a corollary of [a/Fe] and [Fe/H]?

[FromMarie's work, we KNOW that there's more information ...]

Plan to find out:
Application of the age label (as is) to the DR12  RC sample, yields <age>([Fe/H],[a/Fe]);

let's define   D(age)_Kepler  == age_Kepler - <age>([Fe/H],[a/Fe])  and
                    D(age)_Cannon  == age_Cannon - <age>([Fe/H],[a/Fe])
and plot

                  D(age)_Kepler vs  D(age)_Cannon  (as opposed to age_Kepler vs age_Cannon),
i.e. we look at a version of the age vs age plot, compressed around the mean age at a given abundance.

If the Cannon teaches us nothing new about ages beyond that which could be known from
[Fe/H] and [a/Fe], we should get a scatter plot.

HWR's conjecture: the extent along the 1-to-1 line will be ~2 times the scatter..; so, there is more in the spectra...

Should Euclid make 2 passes around the sky?

At the moment, Euclid's wide (15k sqdeg) survey will take all of its data on a given patch
of the sky in one pass (few hours), with some field overlap at the edges.

If the overhead penalty is defensible, it may be conceivable to make two passes around
the sky, taking half the exposure time at each, with the passes separated by ~1 year.

One of the things to think about is which aspects of the time-domain science are enabled by this?

Supernovae? Proper motions of very nearby, very red objects? AGN?


Steps (HWR) for June meeting:

take Mark Cropper's draft of alternate survey strategies, and get Bob Scaramella's
documentation on the survey strategy.

Get Jason involved (weak lensing), Bob Nichol (large scale structure),

Goal:  -- provide the scope document
-- lay out which impact needs to be to be considered:

-- what is the impact of the alternate strategies on the
    basic area-S/N- considerations

-- impact on data reduction, data release policy

-- impact on weak-lensing and LSS science