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:20211207T055414Z
LOCATION:Online
DTSTART;TZID=America/Chicago:20211119T103000
DTEND;TZID=America/Chicago:20211119T105000
UID:submissions.supercomputing.org_SC21_sess338_ws_xloop101@linklings.com
SUMMARY:High-Performance Hybrid-Global-Deflated-Local Optimization with Ap
 plications to Active Learning
DESCRIPTION:Workshop\n\nHigh-Performance Hybrid-Global-Deflated-Local Opti
 mization with Applications to Active Learning\n\nNoack\n\nMathematical opt
 imization lies at the core of many science and industry applications. One 
 important issue with many current optimization strategies is a well-known 
 trade-off between the number of function evaluations and the probability t
 o find the global, or at least sufficiently high-quality local optima. In 
 machine learning (ML), and by extension in active learning --- for instanc
 e for autonomous experimentation --- mathematical optimization is often us
 ed to find the underlying uncertain surrogate model from which subsequent 
 decisions are made and therefore ML relies on high-quality optima to obtai
 n the most accurate models. Active learning often has the added complexity
  of missing offline training data; therefore, the training has to be condu
 cted during the data collection which can stall the acquisition if standar
 d methods are used.   \n\nIn this work, we highlight recent efforts to cre
 ate a high-performance hybrid optimization algorithm (HGDL), combining der
 ivative-free global optimization strategies with local, derivative-based o
 ptimization, ultimately yielding an ordered list of unique local optima. R
 edundancies are avoided by deflating the objective function around earlier
  encountered optima. HGDL is designed to take full advantage of parallelis
 m by having the most computationally expensive process, the local first an
 d second-order-derivative-based optimizations, run in parallel on separate
  compute nodes in separate processes. In addition, the algorithm runs asyn
 chronously; as soon as the first solution is found, it can be used while t
 he algorithm continues to find more solutions.  We apply the proposed opti
 mization and training strategy to Gaussian-Process-driven stochastic funct
 ion approximation and active learning.\n\nTag: Online Only, Applications, 
 Scientific Computing\n\nRegistration Category: Workshop Reg Pass
END:VEVENT
END:VCALENDAR