Super network model, deep learning model training method and information recommendation method

• Main Authors: DONG DAXIANG, WU SHIGUANG, JING QINGHE, WANG YAQING
• Format: Patent
• Language: Chinese; English
• Published: 26.05.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention provides a super network model training method, and relates to the technical field of artificial intelligence and network structure search. According to the specific implementation scheme, initial features of a training sample are input into an adaptive network, and adaptive features of the training sample are generated; the initial features of a test sample are input into the prediction network, hidden features of the test sample are generated, the test sample is provided with a label, and the label represents test interaction data between the test user data and the test object data; adjusting the hidden features based on the adaptive features using an adjustment network; determining an output result of the hypernetwork model according to the adjusted hidden features, wherein the output result represents prediction interaction data between the test user data and the test object data; determining the loss of the super network model according to the test interaction data and the prediction intera