BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Chicago
X-LIC-LOCATION:America/Chicago
BEGIN:DAYLIGHT
TZOFFSETFROM:-0600
TZOFFSETTO:-0500
TZNAME:CDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0500
TZOFFSETTO:-0600
TZNAME:CST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260522T150118Z
LOCATION:C146
DTSTART;TZID=America/Chicago:20181113T133000
DTEND;TZID=America/Chicago:20181113T140000
UID:submissions.supercomputing.org_SC18_sess178_pap171@linklings.com
SUMMARY:Large-Scale Hierarchical K-Means for Heterogeneous Many-Core Super
 computers
DESCRIPTION:Liandeng Li (Tsinghua University; National Supercomputing Cent
 er, Wuxi); Teng Yu (University of St Andrews); Wenlai Zhao and Haohuan Fu 
 (Tsinghua University; National Supercomputing Center, Wuxi); Chenyu Wang (
 University of St Andrews; National Supercomputing Center, Wuxi); Li Tan (B
 eijing Technology and Business University); Guangwen Yang (Tsinghua Univer
 sity; National Supercomputing Center, Wuxi); and John Thomson (University 
 of St Andrews)\n\nThis paper presents a novel design and implementation of
  k-means clustering algorithm targeting the Sunway TaihuLight supercompute
 r. We introduce a multi-level parallel partition approach that not only pa
 rtitions by dataflow and centroid, but also by dimension. Our multi-level 
 (nkd) approach unlocks the potential of the hierarchical parallelism in th
 e SW26010 heterogeneous many-core processor and the system architecture of
  the supercomputer. \n\nOur design is able to process large-scale clusteri
 ng problems with up to 196,608 dimensions and over 160,000 targeting centr
 oids, while maintaining high performance and high scalability, significant
 ly improving the capability of k-means over previous approaches. The evalu
 ation shows our implementation achieves performance of less than 18 second
 s per iteration for a large-scale clustering case with 196,608 data dimens
 ions and 2,000 centroids by applying 4,096 nodes (1,064,496 cores) in para
 llel, making k-means a more feasible solution for complex scenarios.\n\nTa
 g: Algorithms, Architectures, Data Analytics, Deep Learning, Networks, Sci
 entific Computing, Visualization\n\nRegistration Category: Tech Program Re
 g Pass\n\nFinalist: BSP Finalist\n\nSession Chair: Tom Peterka (Argonne Na
 tional Laboratory (ANL))\n\n
END:VEVENT
END:VCALENDAR