Wednesday, August 31, 2011

Day 72

Wednesday.  Dinner at Laura & Robert's tonight.


UCB CDI Seminar
Transfer and active learning in astronomical datasets
Ricardo Vilalta


Machine learning in astronomy: (1) "transfer learning" -- exploit the existence of a data model from a similar (but not exactly the same) domain of application to the new domain.  (2) "active learning" -- dynamically select only those instances of the new domain most informative for prediction.  Both paradigm can greatly enhance the predictive accuracy and efficiency of learning.  Application: automatic geomorphic mapping of planet Mars.  Also: variable star classification, galaxy classification, etc.


1108.5384
The impact of baryon physics on the structure of high-redshift galaxies
Zemp, Gnedin, Gnedin, Kravtsov


Study the structure of galaxies at z>2 using cosmo simulations with improved modeling of the ISM and star formation.  Stellar distribution is compact because H_2 (where stars form) is more concentrated towards the center of galaxies than atomic H.  For halos above 1e11 M_sun, the median half-mass radius of stellar disks is 0.8 kpc at z=3.  Vertical structure of molecular disk is much thinner than that of the atomic neutral gas (why?).  When radiative transfer included (molecular dynamics, I presume), the DM halo change shape from prolate to mildly oblate (affected by baryon disk?) , but no evidence of a significant DM disk around stellar disk.  Outer halo regions retain orientation acquired during accretion and mergers, and are significantly misaligned with the inner regions.  Radial profile of DM halos contracts in response to baryon dissipation, and establish an approximately isothermal profile throughout most of the halo (!).  Effect accurately described by a modified model of "halo contraction".  The angular momentum of a fixed amount of inner dark matter is approximately conserved over time (?), compared to the dissipationless case where most of it is transferred outward during mergers.  Conservation of the dark matter angular momentum provides supporting evidence for the validity of the halo contraction model in hierarchical galaxy formation processes.


* pretty interesting.  Does it simulate disk/elliptical/dwarf galaxies well?
* supports halo contraction due to baryonic dissipation.


LBL INPA
Distribution function approach to redshift space distortions: N-body simulations
Teppei Okumura


Galaxy redshift surveys: redshift space distortions directly probe the cosmic growth history of density perturbations.  A distribution function approach: RSD written as sum over density-weighted velocity moment correlators.  From N-body simulation, this particular approach show that this formalism predicts the true power spectrum up to sufficiently small scales.  Can also be used for biased objects, such as DM halos and galaxies.


UCB SSL Colloquium
3D simulations of solar surface: from limb to secular models
Laurent Piau


* Kepler mission: A search for habitable planets orbiting other stars, using light curves.  Launched 2009, planned mission lifetime of at least 3.5 years.  
* Johannes Kepler: German mathematician, astronomer and astrologer, b.1571 d.1630.  Assistant to Tycho Brahe.  

No comments:

Post a Comment