From a787fd51d0728d2f83c95eab3b017bfdda5f19d9 Mon Sep 17 00:00:00 2001 From: Matt Bendel Date: Fri, 1 Dec 2023 10:21:34 -0500 Subject: [PATCH] Convert P to latex --- index.html | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/index.html b/index.html index 7fca2e7..052fcda 100644 --- a/index.html +++ b/index.html @@ -238,9 +238,9 @@
Ours

Adapting the Weight on the STD Reward

For the independent-Gaussian case, we derive the correct weight \(\beta\) on the STD reward in closed form. For practical datasets, we propose a method to learn \(\beta\). - In particular, we adapt \(\beta\) during training so that the SNR gain from averaging P posterior samples matches the expected theoretical behavior, the latter of which we derive in our paper. + In particular, we adapt \(\beta\) during training so that the SNR gain from averaging \(P\) posterior samples matches the expected theoretical behavior, the latter of which we derive in our paper.

- The orange curves below show the empirical SNR versus P for various values of \(\beta\) (for the case of multicoil MR image recovery) while the blue dashed curves show the expected theoretical behavior. + The orange curves below show the empirical SNR versus \(P\) for various values of \(\beta\) (for the case of multicoil MR image recovery) while the blue dashed curves show the expected theoretical behavior. The figures show that, for this application, the optimal \(\beta=0.53\). Please see our poster or our paper!