Presentation
Taming optimization variance in compact neural shading networks
DescriptionWe present a training strategy for small networks that improves stability, reducing run-to-run variance by 88% and average loss by 38%. Multiple instances train in parallel on different batch slices, while weaker ones are periodically removed until only the best-performing instance remains, improving reliability for neural shading models.
Event Type
Technical Paper
TimeWednesday, 22 July 20262:20pm - 2:30pm PDT
LocationRoom 408 A
Digital Library
PDF
Session Time & Location
Sunday, 19 July 20266:00pm - 8:45pm PDTHall K
Wednesday, 22 July 20262:00pm - 3:30pm PDTRoom 408 A
Artificial Intelligence/Machine Learning
Modeling
Real-Time
Rendering
Full Conference Supporter
Full Conference
