Sabyasachi Mukherjee, Sayan Mukherjee, Binh-Son Hua, Nobuyuki Umetani, Daniel Meister, August 2021 We consider the problem of accelerating sequences of Monte-Carlo estimates of a finite integral, which has a slow convergence rate. We achieve faster convergence with the help of sequence transformation: by changing the sequence of Monte-Carlo estimates into another sequence that converges faster. Demonstrating the difficulties of doing this analytically, we use a data-driven approach that works well for several 1D integrals as well as scenes in light transport simulation. <a href="https://homepages.math.uic.edu/~potla/papers/neuralSeqTr.pdf" rel="noopener noreferrer" target="_blank" style="color: rgb(2, 79, 159); background-color: rgb(255, 255, 255);">Pacific Graphics 2021</a>

Neural Sequence Transformation, August 2021

blueqat research