BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:America/New_York BEGIN:DAYLIGHT DTSTART:20240310T030000 TZOFFSETFROM:-0500 TZOFFSETTO:-0400 RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3 TZNAME:EDT END:DAYLIGHT BEGIN:STANDARD DTSTART:20231105T010000 TZOFFSETFROM:-0400 TZOFFSETTO:-0500 RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11 TZNAME:EST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20240229T234908Z UID:7F1296FA-A2E0-4C7D-96B4-6EE45FF28333 DTSTART;TZID=America/New_York:20240208T180000 DTEND;TZID=America/New_York:20240208T190000 DESCRIPTION:Bayesian particle flow has been used for deep learning\, nonlin ear filtering\, weather forecasting\, medical imaging and many other appli cations. This algorithm is many orders of magnitude faster than standard p article filters\, and it is embarrassingly parallelizable\, unlike standar d particle filters. However\, it can be extremely stiff and hence difficul t to design.\n\nWe derive a new theory of Bayesian particle flow that bull et proofs the algorithm against stiffness. This generalizes our recent wor k (Dai & Daum\, IEEE AESS Transactions June 2023). We also derive a very s imple upper bound on the stiffness of our previous algorithm that shows th at the stiffness is infinite if we do not measure all components of the st ate vector. The new theory fixes this problem completely.\n\nMany research ers have attempted to apply our previous particle flow algorithms\, but so metimes they have gotten disappointing results owing to stiffness. Such re searchers might be experts in fancy estimation algorithms\, but they are n ot experts in mitigating stiffness for Ito stochastic differential equatio ns. We solve this problem by bullet-proofing the particle flow algorithm i tself against stiffness\, rather than using stiff ODE solvers that require a large amount of computer run time\, and which are not parallelizable. W e give a simple intuitive interpretation of the new algorithm.\n\nThis tal k is for normal engineers who do not have Bayesian algorithms for breakfas t.\n\nPlease register below if you would like to receive any eventual last minute notification regarding this event.\n\nSpeaker(s): Fred Daum – Pr incipal Fellow at Raytheon Corporation\, \n\nAgenda: \n5:45pm - Webex link live\n\n6-7pm - Presentation\n\n7-7:15 - Questions and Discussion\n\nVirt ual: https://events.vtools.ieee.org/m/401454 LOCATION:Virtual: https://events.vtools.ieee.org/m/401454 ORGANIZER:menders@ieee.org SEQUENCE:10 SUMMARY:Bullet-proofing particle flow filters against stiffness URL;VALUE=URI:https://events.vtools.ieee.org/m/401454 X-ALT-DESC:Description: <br /><p>Bayesian particle flow has been used for d eep learning\, nonlinear filtering\, weather forecasting\, medical imaging and many other applications.&nbsp\; This algorithm is many orders of magn itude faster than standard particle filters\, and it is embarrassingly par allelizable\, unlike standard particle filters.&nbsp\; However\, it can be extremely stiff and hence difficult to design.</p>\n<p>We derive a new th eory of Bayesian particle flow that bullet proofs the algorithm against st iffness.&nbsp\; This generalizes our recent work (Dai &amp\; Daum\, IEEE A ESS Transactions June 2023).&nbsp\; We also derive a very simple upper bou nd on the stiffness of our previous algorithm that shows that the stiffnes s is infinite if we do not measure all components of the state vector.&nbs p\; The new theory fixes this problem completely.</p>\n<p>Many researchers have attempted to apply our previous particle flow algorithms\, but somet imes they have gotten disappointing results owing to stiffness.&nbsp\; Suc h researchers might be experts in fancy estimation algorithms\, but they a re not experts in mitigating stiffness for Ito stochastic differential equ ations.&nbsp\; We solve this problem by bullet-proofing the particle flow algorithm itself against stiffness\, rather than using stiff ODE solvers t hat require a large amount of computer run time\, and which are not parall elizable.&nbsp\; We give a simple intuitive interpretation of the new algo rithm.</p>\n<p><strong>This talk is for normal engineers who do not have B ayesian algorithms for breakfast.</strong></p>\n<p>Please register below i f you would like to receive any eventual last minute notification regardin g this event.</p><br /><br />Agenda: <br /><p>5:45pm - Webex link live</p> \n<p><strong>6-7pm - Presentation</strong></p>\n<p>7-7:15 - Questions and Discussion</p> END:VEVENT END:VCALENDAR