Meta-learning method and device, initialization method, computing equipment and storage medium

Abstract

The invention relates to a meta-learning method and device of a risk prediction model, an initialization method, computing equipment and a computer readable storage medium. The meta-learning method includes: generating a training task set including a plurality of training tasks, wherein the plurality of training tasks are provided with respective different category predictors; initializing networkweights of a meta-learner, a feature extractor and a task discriminator, the category predictor, the meta-learner, the feature extractor and the task discriminator being an artificial neural network,and the category predictor having the same network structure as the meta-learner; dividing the training tasks in the training task set into a plurality of batches, and updating the network weights ofthe meta-learner, the feature extractor and the task discriminator on the basis of each batch, the updating being carried out according to the category prediction loss and the task discrimination loss. According to the method, the generalization ability of the meta-learner can be improved, so that an optimal risk prediction model can be quickly obtained in small sample training in a financial risk control scene.

Description

technical field [0001] The present invention relates to the technical field of machine learning, in particular to a meta-learning method and device, an initialization method, a computing device, and a computer-readable storage medium. Background technique [0002] Machine learning, especially deep learning, has been successfully applied in many fields such as computer vision, natural language processing, and data mining since its rise. An important factor for its good performance in these fields is the ease of obtaining massive labeled data in these fields. However, in financial risk control scenarios, such as risk control in financial business links such as payment, lending, and wealth management, the data distribution of different customer groups varies greatly, and there are many customer groups with small sample characteristics. Therefore, it is often difficult to collect enough labeled samples for traditional machine learning to extract risk control-related pattern fea...

AI Technical Summary

Problems solved by technology

Therefore, it is often difficult to collect enough labeled samples for traditional machine learning to extract risk control-related pattern features from the data, which is prone to model overfitting

Images