BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Tokyo
X-LIC-LOCATION:Asia/Tokyo
BEGIN:STANDARD
TZOFFSETFROM:+0900
TZOFFSETTO:+0900
TZNAME:JST
DTSTART:18871231T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260817T171531Z
LOCATION:G610\, G Block\, Level 6
DTSTART;TZID=Asia/Tokyo:20241203T130000
DTEND;TZID=Asia/Tokyo:20241203T144500
UID:siggraphasia_SIGGRAPH Asia 2024_sess210_crs_119@linklings.com
SUMMARY:MCMC: Bridging Rendering, Optimization and Generative AI
DESCRIPTION:Gurprit Singh and Gurprit Singh (Max Planck Institute for Info
 rmatics) and Wenzel Jakob (EPFL)\n\nGenerative artificial intelligence (AI
 ) has made unprecedented advances in vision language models over the past 
 two years. These advances are largely due to diffusion-based generative mo
 dels, which are very stable and simple to train.  These diffusion models a
 re tasked to learn the underlying unknown distribution of the training dat
 a samples. During the generative process, new samples (images) are generat
 ed from this unknown high-dimensional distribution. Markov Chain Monte Car
 lo (MCMC) methods are particularly effective in drawing samples from compl
 ex, high-dimensional distributions. This makes MCMC methods an integral co
 mponent for both the training and sampling phases of these models, ensurin
 g accurate sample generation.\n    \n    Gradient-based optimization is at
  the core of modern generative models. The update step during the optimiza
 tion forms a Markov chain where the new update depends only on the current
  state. This allows exploration of the parameter space in a memoryless man
 ner, thus combining the benefits of gradient-based optimization and MCMC s
 ampling. MCMC methods have shown an equally important role in physically b
 ased rendering where complex light paths are otherwise quite challenging t
 o sample from simple importance sampling techniques. \n    \n    A lot of 
 research is dedicated towards bringing physical realism to samples (images
 ) generated from diffusion-based generative models in a data-driven manner
 , however, a unified framework connecting these techniques is still missin
 g. In this course, we take the first steps toward understanding each of th
 ese components and exploring how MCMC could potentially serve as a bridge,
  linking these closely related areas of research. Our course aims to provi
 de necessary theoretical and practical tools to guide students, researcher
 s and practitioners towards the common goal of generative physically based
  rendering.\n\nRegistration Category: Full Access, Full Access Supporter\n
 \nLanguage Format: English Language\n\n
URL:https://asia.siggraph.org/2024/program/?id=crs_119&sess=sess210
END:VEVENT
END:VCALENDAR