Partial mark learning method based on subspace representation and global disambiguation method

Abstract

The invention provides a partial mark learning method based on subspace representation and a global disambiguation method. The method comprises the following steps: constructing a feature matrix and acandidate mark matrix; constructing a feature subspace learning model and a mark global disambiguation model based on the constructed feature matrix and the candidate mark matrix; integrating the feature subspace learning model and the marked global disambiguation model to obtain a hybrid model, and solving the hybrid model by adopting an alternating optimization method to obtain a multi-classification model, a mapping matrix and a partial mark confidence coefficient matrix; and classifying the unseen examples according to the multi-classification model and the mapping matrix, calculating a plurality of mark values of the unseen examples, and determining the mark corresponding to the mark value with the highest prediction confidence as the mark category to which the unseen examples belong. According to the method, the feature subspace representation method and the mark global disambiguation method can be utilized at the same time, the partial mark learning problem is solved from the two aspects of features and marks, and the obtained features have higher representation capacity; and the generated mark confidence coefficient matrix has a better disambiguation effect.

Description

technical field [0001] The invention relates to the field of computer application technology, in particular to a partial label learning method based on subspace representation and global disambiguation method. Background technique [0002] Partially labeled learning can be viewed as a weakly supervised learning framework that aims to learn a multi-classification model from samples with candidate label sets. This learning framework has a wide range of applications in the real world, such as: automatic labeling system, people from different backgrounds made different labels, but only one of the labels was correctly labeled; a news report appeared with multiple names and For a group photo, we need to match the name of the person to the face in the photo. Existing partial label learning methods can be divided into three categories, average disambiguation learning strategies, discriminative disambiguation learning strategies and non-disambiguation learning strategies. [0003] ...

AI Technical Summary

Problems solved by technology

[0006] 1) The disambiguation strategy disambiguates each training example separately, ignoring the global label semantic information of partial label learning, and the real label obtained by disambiguation needs to be further improved;

[0007] 2) The existing partial label learning strategy tends to directly use the original feature space to learn, but in high-dimensional data, redundant features will inevitably be mixed in the original data, which will not only increase the time and space overhead of the training process, And it will reduce the generalization performance of the model

Images