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:20260202T201248Z
LOCATION:Second Floor Atrium
DTSTART;TZID=America/Chicago:20251120T080000
DTEND;TZID=America/Chicago:20251120T170000
UID:submissions.supercomputing.org_SC25_sess533_post293@linklings.com
SUMMARY:CUR-MoE: Portable Mixture-of-Experts with Interpretable High-Ratio
  Compression
DESCRIPTION:Ritesh Bhirud (University of Massachusetts Amherst, Massachuse
 tts Institute of Technology (MIT))\n\nMixture-of-experts (MoE) architectur
 es enable trillion-parameter models but face prohibitive memory scaling, l
 imited compression interpretability, and vendor-specific implementations h
 indering heterogeneous HPC deployment.\n\nWe present the first Julia-based
  MoE framework introducing CUR decomposition for interpretable expert comp
 ression—a novel approach applying CUR matrix factorization to MoE architec
 tures—with hardware-agnostic design. While SVD-based methods provide effec
 tive compression, CUR-MoE offers comparable performance with enhanced inte
 rpretability through preserved column/row structure, maintaining viability
  at high compression ratios (35.29 perplexity at 70% compression). Compreh
 ensive gating evaluation reveals ExpertChoice achieves optimal load balanc
 ing. Julia's LLVM compilation enables consistent 5-6× GPU acceleration acr
 oss NVIDIA, AMD, Intel, and Apple hardware.\n\nOur core implementation is 
 completed, validated on WikiText-2 across platforms. We are expanding comp
 rehensive platform support for Apple Metal and Intel Arc while extending T
 ransformers.jl and Flux.jl integrations. The poster will include visual co
 mparisons, cross-vendor benchmarks, detailed oral explanations, and QR cod
 es with live interactive GitHub examples demonstrating CUR structure prese
 rvation.\n\nTag: Research & ACM SRC Posters\n\nRegistration Category: Tech
 nical Program Reg Pass\n\nSession Chairs: Kento Sato (RIKEN Center for Com
 putational Science (R-CCS)); Chris Schlipalius (Pawsey Supercomputing Rese
 arch Centre; Commonwealth Scientific and Industrial Research Organisation 
 (CSIRO), Australia); and Anja Gerbes (Georg-August-Universität Göttingen)\
 n\n
END:VEVENT
END:VCALENDAR