US 12,346,813 B2
Online knowledge distillation for multi-task learning system, method, device, and program
Geethu Jacob, Karnataka (IN); Vishal Agarwal, Karnataka (IN); and Bjorn Stenger, Tokyo (JP)
Assigned to Rakuten Group, Inc., Tokyo (JP)
Filed by Rakuten Group, Inc., Tokyo (JP)
Filed on Jul. 28, 2022, as Appl. No. 17/875,734.
Prior Publication US 2024/0037386 A1, Feb. 1, 2024
Int. Cl. G06N 3/08 (2023.01); G06T 3/4046 (2024.01)
CPC G06N 3/08 (2013.01) [G06T 3/4046 (2013.01)] 19 Claims
OG exemplary drawing
 
1. A method for image processing using a multi-task neural network framework, the method being executed by one or more processors, the method comprising:
receiving an input image;
performing an image processing task based on the input image using the multi-task neural network framework, wherein the multi-task neural network framework is trained using an adaptive feature distillation function, and wherein the adaptive feature distillation function is based on a comparison of intermediate features of the multi-task neural network framework and intermediate features of a plurality of single-task neural network models; and
generating an output of the image processing task based on up sampling an output of the multi-task neural network framework,
wherein the multi-task neural network framework comprises a shared transformer backbone network and a plurality of task specific heads.