Multi-classification model training method of adaptive weighting loss function

• Main Authors: HE ZHENHAO, JIANG WENCHUN, ZHANG ZHISONG, WANG RONGDI, ZUO HAIQIANG, CAO HUAIXIANG, ZHU JIANCHAO
• Format: Patent
• Language: Chinese; English
• Published: 25.04.2023
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention discloses a multi-classification model training method of a self-adaptive weighting loss function, and belongs to the field of machine learning. The method is characterized by comprising the following steps: step a, establishing a multi-classification model needing to be trained; b, reading data and labels in the data set, inputting the data into the multi-classification model, and calculating the error rate of each category of the model; c, multiplying the loss function calculation formula of each category by the respective loss function weight to obtain total loss; d, carrying out back propagation, and updating model parameters; and e, judging whether the multi-classification model is converged or not, if so, ending, otherwise, returning to the step b. In the multi-classification model training method of the self-adaptive weighting loss function, self-adaptive weighting is performed on different sample weights, manual parameter adjustment and optimization on the weight of the loss function are