Semi-supervised classification method of modified clustering assumption combined with pairwise constraints

Abstract

The invention discloses semi-supervised classification method of modified clustering assumption combined with pairwise constraints. The method relates to a semi-supervised learning algorithm, and includes the steps of: initializing class membership degrees of unlabeled samples through an FCM method; selecting appropriate parameters of lambda1 and lambda2, and calculating initialized alpha, a membership degree function of v(x) and a new objective function of M according to formulas; judging whether an iteration termination condition is reached; if yes, returning the membership degree function v(x), and obtaining a classification decision function of f(x) according to the alpha; and otherwise, re-calculating the initialized alpha, the membership degree function of v(x) and the new objectivefunction of M, and carrying out judgement. According to the method, exploration of modified clustering assumption on the unlabeled samples and utilization of the pairwise constraints on supervisory information are combined to jointly form more completed empirical risk terms, thus knowledge contained by the supervisory information is further mined, and the purpose of algorithm performance improvement is achieved. The method has higher validity and correctness.

Description

technical field [0001] The invention relates to a semi-supervised learning algorithm, in particular to a semi-supervised classification method with modified clustering assumptions and joint pairwise constraints. Background technique [0002] Semi-supervised learning is a learning method between supervised learning and unsupervised learning. The basic premise of learning is: in addition to a large number of unlabeled samples, the labeled samples also provide supervised information such as class labels; semi-supervised learning Supervised learning differs from supervised learning in that it augments the training dataset with a large number of unlabeled samples. The main method of semi-supervised learning is from the perspective of supervised learning. When the labeled samples with supervised information are not enough to train a good model, how to automatically use the information of a large number of unlabeled samples to help improve the performance of the classifier. [000...

AI Technical Summary

Problems solved by technology

[0004] The core idea of ​​the semi-supervised learning method is how to use the knowledge contained in a small number of labeled samples and a large number of unlabeled samples to improve the learning ability of the algorithm. The current mainstream semi-supervised learning algorithm mainly obtains knowledge from unlabeled samples to mine the distribution of data. Information improves the performance of the classifier, but ignores the deep utilization of supervised information such as labeled samples. To a certain extent, the important information contained in the labeled samples is lost, and the knowledge is not maximized. Lack of effectiveness and correctness , the performance of the algorithm is low

For example, an improved idea of ​​clustering hypothesis, which modifies the clustering hypothesis by introducing the concept of membership degree, and improves the usual clustering hypothesis, that is, samples in the same cluster are more likely to have the same class label, to be in The samples in the same cluster have similar membership degrees, and on this basis, a new semi-supervised classification method - semi-supervised classification method based on class membership (SSCCM), but it can be seen that the SSCCM algorithm as a A new semi-supervised classification method, which mainly relies on revised clustering assumptions, and does not make use of supervised information

Images