Semi-supervised extreme learning machine classification method with safety mechanism

Abstract

Semi-supervised learning can utilize marked and unmarked samples to improve performance of classifiers. One of the keys is how to efficiently and safely mine information hidden in the unmarked samples. The invention provides a semi-supervised extreme learning machine classification method with a safety mechanism. The semi-supervised extreme learning machine classification method comprises the steps of: firstly, adopting an extreme learning machine algorithm and a semi-supervised extreme learning machine algorithm to predict probability distribution vectors and category labels of the unmarked samples separately; secondly, adopting a Wasserstein distance to measure a risk degree of the unmarked samples; thirdly, adding a new risk item into an objective function of the semi-supervised extremelearning machine algorithm; and finally, realizing safe semi-supervised multi-class classification. The semi-supervised extreme learning machine classification method can effectively solve the problem that performance of the classifier can be damaged by adding the risky unmarked samples into the semi-supervised extreme learning machine algorithm, and has broad application prospects in the fieldsof computer vision, image identification and brain-computer interfaces.

Description

technical field [0001] The invention belongs to the field of pattern recognition and relates to a semi-supervised ultra-limit learning machine classification method with a safety mechanism. Background technique [0002] Semi-supervised learning has become a research hotspot in the field of machine learning in recent years, and has been extended to many real-life application scenarios, such as spam interception, target detection and tracking, face recognition, etc. Semi-supervised learning is to use both labeled and unlabeled samples to improve the performance of the classifier. The key to semi-supervised learning is how to effectively use the hidden information in unlabeled samples. The commonly used assumptions are: smoothness assumption, clustering assumption, flow Formal assumptions, etc. Among them, the manifold hypothesis has become a research hotspot in recent years because it can discover the manifold structure information between labeled and unlabeled samples. At p...

AI Technical Summary

Problems solved by technology

[0005] To sum up, the security semi-supervised methods mentioned above are all developed on the basis of SVM and KMSE methods, but how to provide a reasonable security semi-supervised mechanism for the currently widely used SS-ELM algorithm , this problem has not been effectively solved

Images