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:20260202T201803Z
LOCATION:275
DTSTART;TZID=America/Chicago:20251120T135200
DTEND;TZID=America/Chicago:20251120T141500
UID:submissions.supercomputing.org_SC25_sess296_pap458@linklings.com
SUMMARY:DPAR: High-Performance, Secure, and Scalable Differential Privacy-
 Based AllReduce
DESCRIPTION:Hao Qi and Weicong Chen (University of California, Merced); Ch
 enghong Wang (Indiana University); and Xiaoyi Lu (University of California
 , Merced)\n\nSecure, efficient, and scalable AllReduce-based data aggregat
 ion is essential for artificial intelligence (AI) and scientific applicati
 ons on modern high performance computing (HPC) and cloud infrastructures. 
 As AllReduce is increasingly used across these distributed infrastructures
 , privacy has become a critical concern. State-of-the-art (SOTA) homomorph
 ic encryption (HE)-based AllReduce solutions introduce high overhead, requ
 ire secure key exchanges, and remain vulnerable to collusion.\n\nWe propos
 e DPAR, the first differentially private, collusion-resistant AllReduce fr
 amework optimized for large-scale HPC and AI workloads. DPAR introduces th
 ree key innovations: integrating differential privacy (DP) to eliminate co
 llusion risks without key exchanges, scalable noise growth to preserve acc
 uracy, and performance optimizations using a noise pooling mechanism. \n\n
 DPAR is a drop-in Message Passing Interface (MPI) AllReduce replacement, p
 roviding strong privacy with minimal performance cost. Evaluated on Delta 
 and Frontier supercomputers with up to 8,192 cores, DPAR outperforms the S
 OTA HE solution by up to 34.7% in modern AI workloads.\n\nTag: Architectur
 es & Networks, HPC for Machine Learning, Performance Measurement, Modeling
 , & Tools, Programming Frameworks\n\nRecording: Livestreamed, Recorded\n\n
 Registration Category: Technical Program Reg Pass\n\nSession Chair: Shaswo
 t Shresthamali (Kyushu University; Keio University, Tokyo)\n\n
END:VEVENT
END:VCALENDAR