BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:On distributed Bayesian computation - Harry van Zanten\, Universit
 y of Amsterdam
DTSTART:20190517T150000Z
DTEND:20190517T160000Z
UID:TALK115936@talks.cam.ac.uk
CONTACT:Dr Sergio Bacallado
DESCRIPTION:Due to the rapidly increasing amount of available information\
 , computer scientists and statisticians are facing new challenges to deal 
 with big data problems. Some of the most popular and frequently applied ap
 proaches to solve this problem are distributed methods where the data is s
 plit and handled by multiple local servers or cores and computations are d
 one locally\, parallel to each other. Then the local machines transmit the
  outcome of their computations to a global server which aggregates the loc
 al results into a global one. \n\nIn the (Bayesian) literature various met
 hods were proposed for distributed computational methods with seemingly go
 od practical performance\, but with limited theoretical underpinning. In o
 ur work we investigate the existing distributed methods in a standard nonp
 arametric setting (the Gaussian signal-in-white-noise model) and compare t
 heir theoretical performance\, i.e. posterior contraction rates and covera
 ge of credible sets. Next we ask what is fundamentally possible in the dis
 tributed setting. To make this precise we add certain communication restri
 ctions and prove minimax lower bounds for distributed procedures under suc
 h restrictions. Moreover\, we exhibit distributed procedure attaining the 
 bounds. Finally\, we address the issue of adaptive distributed estimation.
 \n\nBased on joint work with Botond Szabo.
LOCATION:MR12
END:VEVENT
END:VCALENDAR