BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Australia/Melbourne
X-LIC-LOCATION:Australia/Melbourne
BEGIN:DAYLIGHT
TZOFFSETFROM:+1000
TZOFFSETTO:+1100
TZNAME:AEDT
DTSTART:19721003T020000
RRULE:FREQ=YEARLY;BYMONTH=4;BYDAY=1SU
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:19721003T020000
TZOFFSETFROM:+1100
TZOFFSETTO:+1000
TZNAME:AEST
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20240214T070245Z
LOCATION:Meeting Room C4.9+C4.10\, Level 4 (Convention Centre)
DTSTART;TZID=Australia/Melbourne:20231213T152000
DTEND;TZID=Australia/Melbourne:20231213T153000
UID:siggraphasia_SIGGRAPH Asia 2023_sess168_papers_906@linklings.com
SUMMARY:Joint Sampling and Optimisation for Inverse Rendering
DESCRIPTION:Technical Papers\n\nMartin Balint, Karol Myszkowski, Hans-Pete
 r Seidel, and Gurprit Singh (Max Planck Institute for Informatics)\n\nWhen
  dealing with difficult inverse problems such as inverse rendering, using 
 Monte Carlo estimated gradients to optimise parameters can slow down conve
 rgence due to variance. Averaging many gradient samples in each iteration 
 reduces this variance trivially. However, for problems that require thousa
 nds of optimisation iterations, the computational cost of this approach ri
 ses quickly.\n\nWe derive a theoretical framework for interleaving samplin
 g and optimisation. We update and reuse past samples with low-variance fin
 ite-difference estimators that describe the change in the estimated gradie
 nts between each iteration. By optimally combining proportional and finite
 -difference samples, we continuously reduce the variance of our novel grad
 ient meta-estimators throughout the optimisation process. We investigate h
 ow our estimator interlinks with Adam and derive a stable combination.\n\n
 We implement our method for inverse path tracing and demonstrate how our e
 stimator speeds up convergence on difficult optimisation tasks.\n\nRegistr
 ation Category: Full Access\n\nSession Chair: Soo-Mi Choi (Sejong Universi
 ty)
URL:https://asia.siggraph.org/2023/full-program?id=papers_906&sess=sess168
END:VEVENT
END:VCALENDAR