[2311.14649] Studying in Deep Issue Graphs with Gaussian Perception Propagation


Obtain a PDF of the paper titled Studying in Deep Issue Graphs with Gaussian Perception Propagation, by Seth Nabarro and a couple of different authors

Obtain PDF
HTML (experimental)

Summary:We suggest an method to do studying in Gaussian issue graphs. We deal with all related portions (inputs, outputs, parameters, latents) as random variables in a graphical mannequin, and examine each coaching and prediction as inference issues with completely different noticed nodes. Our experiments present that these issues might be effectively solved with perception propagation (BP), whose updates are inherently native, presenting thrilling alternatives for distributed and asynchronous coaching. Our method might be scaled to deep networks and offers a pure means to do continuous studying: use the BP-estimated parameter marginals of the present job as parameter priors for the subsequent. On a video denoising job we exhibit the good thing about learnable parameters over a classical issue graph method and we present encouraging efficiency of deep issue graphs for continuous picture classification.

Submission historical past

From: Seth Nabarro [view email]
Fri, 24 Nov 2023 18:31:11 UTC (3,272 KB)
Wed, 28 Feb 2024 15:56:48 UTC (3,075 KB)

Supply hyperlink


Please enter your comment!
Please enter your name here