Active learning method based on generative adversarial model

Abstract

The invention discloses an active learning method based on a generative adversarial model. The method comprises the following steps: firstly, training a generative adversarial model by using a labeleddata set and an unlabeled data set, and converting each sample in the labeled data set and the unlabeled data set into a vector in a certain hidden feature space by using the trained generative adversarial model; then, training a reference classifier by utilizing the conversion vector of the labeled data set; testing the samples in the test set by using the trained reference classifier, and checking whether a preset termination condition is reached or not; if the preset termination condition is met, selecting samples around the decision boundary of the real reference classifier from the unlabeled data set; then, enabling a labeler to label the category of the selected sample, and adding the labeled data set; and repeating the above steps until a preset termination condition is satisfied.According to the method, the cost of manually labeling samples can be reduced, and the high-performance model is obtained by training at the labeling cost as low as possible.

Description

technical field [0001] The invention relates to the field of computer technology, in particular to an active learning method based on a generated confrontation model. Background technique [0002] At present, deep learning has achieved remarkable success in many fields, such as computer vision, speech recognition, natural language processing and so on. Despite the general success of neural networks in so many tasks, it has an obvious disadvantage, neural networks require a large amount of labeled data in order to learn a large number of parameters to ensure the performance of the model. And having more data is almost always better, and the performance of a neural network usually does not saturate as the dataset increases, but instead improves its generalization performance. [0003] From an algorithmic perspective, one would expect to have more labeled data. But in fact, labeling data sets requires a lot of time and energy, especially in professional fields such as medical...

AI Technical Summary

Problems solved by technology

[0007] In view of this, the present invention provides an active learning method based on a generative confrontation model, which can use the generative confrontation model to select samples around the decision boundary of the real model, and the selected samples have a high contribution to the improvement of model performance, and at the same time Solved the problems of selection to redundant samples, selection to isolated sample points, etc.

Images