公开(公告)号：US20240160934A1

公开(公告)日：2024-05-16

申请号：US18450571

申请日：2023-08-16

Applicant: MEDIATEK INC.

Inventor： Hao CHEN ,  Po-Hsiang YU ,  Yu-Cheng LO ,  Cheng-Yu YANG ,  Peng-Wen CHEN

IPC: G06N3/082

CPC classification number: G06N3/082

Abstract: A method for removing branches from trained deep learning models is provided. The method includes steps (i)-(v). In step (i), a trained model is obtained. The trained model has a branch structure involving one or more original convolutional layers and a shortcut connection. In step (ii), the shortcut connection is removed from the branch structure. In step (iii), a reparameterization model is built by linearly expanding each of the original convolutional layers into a reparameterization block in the reparameterization model. In step (iv), parameters of the reparameterization blocks are optimized by training the reparameterization model. In step (v), each of the optimized reparameterization blocks is transformed into a reparameterized convolutional layer to form a branchless structure that replaces the branch structure in the trained model.