BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20260523T080442EDT-6211Av1uLd@132.216.98.100
DTSTAMP:20260523T120442Z
DESCRIPTION:Abstract\n\nRapid increase in computational power have propelle
 d both computer graphics and machine learning toward achieving increasingl
 y complex objectives. The two fields have experienced a cross-pollination 
 of ideas and technical expertise\, advancing them collectively. Neural gra
 phics explores the application of machine learning techniques in various a
 spects of computer graphics\, including rendering\, animation\, view synth
 esis\, and other visual tasks. It utilizes learned techniques to enhance t
 raditional computer generated imagery (CGI) and visual content creation\, 
 while also enabling new pipelines for non-experts to create and consume im
 mersive graphics experiences. The aim of neural graphics is to harness the
  power of data-driven models for enhancing bidirectional interactions betw
 een simulations and the real-world across various applications\, from vide
 o games and virtual reality to computer aided design and digital art. Thes
 e applications require efficient algorithms and paradigms to address their
  ever increasing computational demands.\n\nEfficiency is crucial for appli
 cations that must operate within strict energy budgets while maintaining d
 esired performance and quality levels. Many real-time graphics application
 s demand precise refresh rates of 60 or 90 Hz\, necessitating the developm
 ent of specially crafted algorithms to meet the performance and efficiency
  targets. Efficient algorithms not only enhance existing applications but 
 also enable new ones that were previously impractical due to energy or per
 formance constraints. At its core\, energy is essential for processing dat
 a through arithmetic manipulations or moving data between storage and proc
 essing elements. Often\, the latter requires significantly more energy\, a
 nd achieving maximum efficiency requires balancing these two aspects. In m
 odern devices\, memory operations are slower than arithmetic operations\, 
 and the scaling of memory subsystems lags behind arithmetic units with eac
 h new hardware generation. Consequently\, more algorithms are becoming bot
 tlenecked due to memory constraints rather than compute limitations. Hence
 \, an efficient algorithm should not only require fewer computations but a
 lso minimize data movement\, known as bandwidth. While neural networks hav
 e played a pivotal role in many graphics applications\, their direct appli
 cation in graphics may lead to inefficiencies due to their substantial ban
 dwidth and compute requirements.\n\nThis dissertation systematically addre
 sses efficiency challenges by taking a bottom-up approach at three levels:
  primitive\, network\, and application levels. Any efficiency improvement 
 at a lower level permeates through the levels above. At the first level\, 
 we introduce a new primitive called \textit{differentiable indirection}\, 
 which can be used to construct more complex networks or be combined with o
 ther neural primitives. Our novel primitive is more compact and intrinsica
 lly more compute and bandwidth efficient compared to other primitives\, su
 ch as multilayer perceptrons or neural fields. It has been tested across v
 arious graphics tasks\, including geometry representation\, shading\, text
 uring\, radiance fields\, and holds potential for applications beyond grap
 hics. Moving to the network level\, we demonstrate how to optimize a \text
 it{convolutional neural network} to minimize bandwidth requirements\, maki
 ng it suitable for real-time graphics applications like shadow synthesis. 
 Finally\, at the application level\, we optimize our pipeline to utilize s
 maller and more efficient networks in the context of shadow synthesis and 
 soft-body animation. For shadow synthesis\, we leverage domain knowledge t
 o fine-tune input features\, ensure more robust training for temporal stab
 ility\, and perform network pruning based on the application environment. 
 In the realm of soft-body simulation\, we employ dimensionality reduction 
 to develop a compact\, reduced-space neural operator for the rapid synthes
 is of latent temporal trajectories.\n
DTSTART:20240726T154500Z
DTEND:20240726T170000Z
LOCATION:Room 603\, McConnell Engineering Building\, CA\, QC\, Montreal\, H
 3A 0E9\, 3480 rue University
SUMMARY:PhD defence of Sayantan Datta – Efficient Neural Graphics
URL:/ece/channels/event/phd-defence-sayantan-datta-eff
 icient-neural-graphics-357998
END:VEVENT
END:VCALENDAR