Semi-supervised multi-label learning method based on SPUM data

Abstract

The invention discloses a semi-supervised multi-label learning method based on SPUM data. The method comprises the following steps of: considering single positive example unlabeled multi-label data (SPUM) data in which only one label in a plurality of labels of data is definitely labeled as positive and the other labels are in an unlabeled state, and solving a globally optimal solution through information transmission when the unlabeled data and the SPUM data are dispersed in different data nodes. According to the method, a real-time label prediction loss function based on positive class number constraint is adopted, so that the algorithm can learn a large amount of information by constraining the positive class number and alternately optimizing label prediction and model output in a scene that only a single label of partial data is labeled as a positive example, and the cost of a labeling task is greatly reduced. In a distributed network in which resources such as communication bandwidth are limited, an iterative strategy based on event triggering is adopted, so that the gradient information amount is allowed to be transmitted after reaching a certain degree, and the network burden is greatly reduced while the performance is kept.

Description

technical field [0001] The present invention relates to a semi-supervised multi-label learning method based on SPUM data, which can learn an algorithm for a good classification model on unlabeled data and multi-label data with only a single positive label marked, and this algorithm can only pass between nodes A small amount of information transfer learns a globally optimal classification model. Background technique [0002] In actual multi-label learning, a data may have multiple labels that need to be predicted. For example, in the field of target detection, there are usually multiple targets in a picture, and a text in text sentiment analysis will also have multiple emotions. The current state-of-the-art multi-label classification algorithms usually require each label of each data to be manually marked with +1 or 0 to indicate the existence or non-existence of the class, but this is usually a cumbersome task so that the complete label data is expensive. If the multi-labe...

AI Technical Summary

Problems solved by technology

The current state-of-the-art multi-label classification algorithms usually require each label of each data to be manually marked with +1 or 0 to indicate the existence or non-existence of the class, but this is usually a cumbersome task so that the complete label data is expensive

Images