BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20250822T115805Z
LOCATION:Campussaal - Plenary Room
DTSTART;TZID=Europe/Stockholm:20250616T102000
DTEND;TZID=Europe/Stockholm:20250616T105000
UID:submissions.pasc-conference.org_PASC25_sess149_pos140@linklings.com
SUMMARY:P03 - Bit-IF: An Incremental Sparse Tensor Format for Maximizing E
 fficiency in Tensor-Vector Multiplications
DESCRIPTION:Xiaohe Niu (mcs Software AG); Georg Meyer (Friedrich-Alexander
 -Universität Erlangen-Nürnberg, Università della Svizzera italiana); Dimos
 thenis Pasadakis (Università della Svizzera italiana, Panua Technologies);
  Albert-Jan N. Yzelman (Huawei Zurich Research Center); and Olaf Schenk (U
 niversità della Svizzera italiana, Panua Technologies)\n\nThis poster pres
 ents **Bit-IF** (Incremental Sparse Fibers with Bit Encoding), a novel spa
 rse tensor format designed to reduce the storage requirements of large ten
 sors and improve the efficiency of tensor operations, particularly of tens
 or-vector multiplication (TVM). As datasets in many scientific fields incr
 ease in dimensionality, size, and sparsity, efficient storage and computat
 ion methods become essential. Current state-of-the-art sparse tensor forma
 ts achieve memory-efficient representations but often require extensive in
 dexing or pre-computation, limiting flexibility and efficiency. Unlike exi
 sting formats, Bit-IF only records index increments encoded by a compact b
 it array. This mode-independent approach allows for an arbitrary index tra
 versal during the TVM. Bit-IF's design characteristics significantly reduc
 e memory overhead, improve data locality, and eliminate the need for multi
 ple tensor copies or mode-specific preprocessing before performing a TVM. 
 Our analysis and initial comparative studies show that Bit-IF reduces memo
 ry consumption and computation time compared to COO-based approaches. Its 
 mode independence and incremental indexing allow for flexible traversal or
 ders, enabling the use of space-filling curves such as Z-curves or Hilbert
  curves to improve data locality and scalability. We plan to extend the ap
 plicability of this method to other tensor operations, such as tensor-matr
 ix and Khatri-Rao products.\n\nSession Chair: Chris Cantwell (Imperial Col
 lege London)\n\n
END:VEVENT
END:VCALENDAR