Method for training deep neural network (DNN) using auxiliary regression targets

US11288567B2 · kind B2 · utility

Assignee

• NEC CORPORATION · JP

Inventors

• Mischa Schmidt · Heidelberg, DE
• Tobias Jacobs · Mannheim, DE

Key dates

• Filing date: Nov 19, 2018
• Grant date: Mar 29, 2022
• Expiry date: Oct 5, 2040

Classification

• Technology area (CPC G): Physics
• CPC primary: G06N3/047
• WIPO field: Computer technology
• WIPO sector: Electrical engineering

Abstract

A method for training a machine learning model includes calculating auxiliary regression targets (ARTs) for a training data set, modifying an input neural network architecture to provide a modified neural network architecture that includes a parallel neural network layer stack for regressing the ARTs, and training the modified neural network architecture on the ARTs in addition to original machine learning problem targets for the training data set.