BEGIN:VCALENDAR
PRODID:-//Microsoft Corporation//Outlook MIMEDIR//EN
VERSION:2.0
BEGIN:VEVENT
DTSTART:20151117T220000Z
DTEND:20151117T223000Z
LOCATION:18AB
DESCRIPTION;ENCODING=QUOTED-PRINTABLE:ABSTRACT: We introduce a general-dimensional, kernel-independent, algebraic=0Afast multipole method and apply it to kernel regression. The=0Amotivation for this work is the approximation of=0Akernel matrices, which appear in mathematical=0Aphysics, approximation theory, non-parametric statistics, and machine=0Alearning. Existing fast multipole methods are asymptotically optimal,=0Abut the underlying constants scale quite badly with the ambient space=0Adimension. We introduce a method that mitigates this shortcoming; it=0Aonly requires kernel evaluations and scales well with the problem=0Asize, the number of processors, and the ambient dimension---as long as=0Athe intrinsic dimension of the dataset is small. We test the=0Aperformance of our method on several synthetic datasets. As a=0Ahighlight, our largest run was on an image dataset with 10 million=0Apoints in 246 dimensions.
SUMMARY:A Kernel-Independent FMM in General Dimensions
PRIORITY:3
END:VEVENT
END:VCALENDAR