BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:America/Denver
X-LIC-LOCATION:America/Denver
BEGIN:DAYLIGHT
TZOFFSETFROM:-0700
TZOFFSETTO:-0600
TZNAME:MDT
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=2SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:-0600
TZOFFSETTO:-0700
TZNAME:MST
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=11;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20240116T185913Z
LOCATION:E Concourse
DTSTART;TZID=America/Denver:20231115T100000
DTEND;TZID=America/Denver:20231115T170000
UID:submissions.supercomputing.org_SC23_sess301_drs123@linklings.com
SUMMARY:I/O Efficient Machine Learning
DESCRIPTION:Doctoral Showcase, Posters\n\nMeghana Madhyastha (Johns Hopkin
 s University, Argonne National Laboratory (ANL))\n\nMy research focuses on
  systems optimizations for machine learning, specifically on I/O efficient
  model storage and retrieval.\n\nThe first part of my work focuses on effi
 cient inference serving of tree ensemble models. Tree structures are inher
 ently not cache friendly and their traversal incurs random I/Os. We develo
 ped two systems - Blockset (Block Aligned Serialized Trees) and T-REX (Tre
 e Rectangles).\n\nBlockset improves inference latency in the scenario wher
 e the model doesn’t fit in memory. It introduces the concept of selective 
 access for tree ensembles in which only the parts of the model needed for 
 inference are deserialized and loaded into memory. It uses principles from
  external memory algorithms to rearrange tree nodes in a block aligned for
 mat to minimize the number of I/Os needed for inference. T-REX optimizes i
 nference latency for both in-memory inference as well as inference when th
 e model doesn’t fit in memory. T-REX reformulates decision tree traversal 
 as hyperrectangle enclosure queries using the fact that decision trees par
 tition the space into convex hyperrectangles.  The test points are then qu
 eried for enclosure inside the hyperrectangles. In doing random I/O is tra
 ded for additional computation.\n\nThe second part of my work focuses on e
 fficient deep learning model storage. We implemented a deep learning model
  repository that requires fine-grained access to individual tensors in mod
 els. This is useful in applications such as transfer learning, where indiv
 idual tensors in layers are transferred from one model to another. We’re c
 urrently working on caching and prefetching popular tensors based on appli
 cation level hints.\n\nTag: Artificial Intelligence/Machine Learning, I/O 
 and File Systems\n\nRegistration Category: Tech Program Reg Pass, Exhibits
  Reg Pass
END:VEVENT
END:VCALENDAR