BEGIN:VCALENDAR
VERSION:2.0
PRODID:IEEE vTools.Events//EN
CALSCALE:GREGORIAN
BEGIN:VTIMEZONE
TZID:America/New_York
BEGIN:DAYLIGHT
DTSTART:20260308T030000
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3
TZNAME:EDT
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20251102T010000
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11
TZNAME:EST
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20251120T200234Z
UID:F7DA4A68-4FDA-4624-B137-A9D5B11ED8F5
DTSTART;TZID=America/New_York:20251120T100000
DTEND;TZID=America/New_York:20251120T113000
DESCRIPTION:Abstract: In many applications\, machine learning involves mana
 ging data across multiple devices without the availability of a central se
 rver\, necessitating a decentralized learning approach. In such settings\,
  nodes are susceptible to failures from malfunctions or cyberattacks\, whi
 ch can undermine traditional learning algorithms. This paper addresses the
  robustification of decentralized learning amidst Byzantine failures\, whe
 re nodes can arbitrarily deviate\, threatening system stability. Prior wor
 ks have utilized ad-hoc methods akin to robust statistics\; however\, we p
 ropose a formal integration of robust statistical principles into the lear
 ning process for a more systematic approach. We introduce BRIDGE\, a scala
 ble Byzantine-resilient decentralized machine learning framework\, designe
 d to fortify resilience and offer structured analysis against Byzantine be
 haviors. BRIDGE comes with algorithmic and statistical convergence guarant
 ees for both strongly convex and select nonconvex problems. Our experiment
 s validate BRIDGE's scalability and effectiveness\, underscoring its robus
 tness and showcasing the benefits of incorporating robust statistics into 
 decentralized learning systems formally.\n\nCo-sponsored by: Department of
  Electrical and Computer Engineering\, University of Michigan-Dearborn\n\n
 Bldg: Engineering Lab Building\, ELB 2390\, University of Michigan-Dearbor
 n\, Dearborn\, Michigan\, United States\, 48128
LOCATION:Bldg: Engineering Lab Building\, ELB 2390\, University of Michigan
 -Dearborn\, Dearborn\, Michigan\, United States\, 48128
ORGANIZER:alihssn@umich.edu
SEQUENCE:10
SUMMARY:Robustness and Resilience in Decentralized Learning
URL;VALUE=URI:https://events.vtools.ieee.org/m/515695
X-ALT-DESC:Description: <br /><p><strong>Abstract:</strong>&nbsp\;In many a
 pplications\, machine learning involves managing data across multiple devi
 ces without the availability of a central server\, necessitating a decentr
 alized learning approach. In such settings\, nodes are susceptible to fail
 ures from malfunctions or cyberattacks\, which can undermine traditional l
 earning algorithms. This paper addresses the robustification of decentrali
 zed learning amidst Byzantine failures\, where nodes can arbitrarily devia
 te\, threatening system stability. Prior works have utilized ad-hoc method
 s akin to robust statistics\; however\, we propose a formal integration of
  robust statistical principles into the learning process for a more system
 atic approach. We introduce BRIDGE\, a scalable Byzantine-resilient decent
 ralized machine learning framework\, designed to fortify resilience and of
 fer structured analysis against Byzantine behaviors. BRIDGE comes with alg
 orithmic and statistical convergence guarantees for both strongly convex a
 nd select nonconvex problems. Our experiments validate BRIDGE's scalabilit
 y and effectiveness\, underscoring its robustness and showcasing the benef
 its of incorporating robust statistics into decentralized learning systems
  formally.<br><br></p>
END:VEVENT
END:VCALENDAR